本次复现的模型为时空注意神经网络(STANet)。 STANet设计了两种类型的自我注意模块。基本时空注意模块(BAM)。金字塔时空注意模块(PAM)
参考实现：https://github.com/justchenhao/STANet
在LEVIR的测试集的测试效果如下表,达到验收指标，F1-Score=0.91 满足精度要求 model.pdiparams文件大小为14.6m model.pdmodel文件大小为5.11m，和为19.7m 满足要求 综上所述 满足挑战指标要求
通过以下命令安装对应依赖
cd STANET_Paddle/
pip install -r requirements.txt
pip install https://paddle-model-ecology.bj.bcebos.com/whl/paddleclas-0.0.0-py3-none-any.whl下载地址:
https://aistudio.baidu.com/aistudio/datasetdetail/136610
数据集下载解压后需要生成.txt文件用于训练。执行以下命令。
#切片
python ./STANET_Paddle/tools/spliter-cd.py --image_folder data/LEVIR-CD --block_size 256 --save_folder dataset参数介绍：
- image_folder:数据集路径
- block_size:切片大小
- save_folder:保存路径
# 创建列表
python ./STANET_Paddle/tools/create_list.py --image_folder ./dataset/train --A A --B B --label label --save_txt train.txt
python ./STANET_Paddle/tools/create_list.py --image_folder ./dataset/val --A A --B B --label label --save_txt val.txt
python ./STANET_Paddle/tools/create_list.py --image_folder ./dataset/test --A A --B B --label label --save_txt test.txt参数介绍：
- image_folder:切片后数据集路径
- -A -B -label :A时相、B时相、label的子路径名
- save_txt:保存名
运行一下命令进行模型训练，在训练过程中会对模型进行评估，启用了VisualDL日志功能，运行之后在/output/stanet/vdl_log 文件夹下找到对应的日志文件
训练过程中的vdl文件为vdlrecords.1655664431.log
!python ./STANET_Paddle/tutorials/train/stanet_train_bonesmall.py --data_dir=./dataset/ --out_dir=./output1/stanet/ --batch_size=8 --num_epoch=100参数介绍：
-
data_dir:数据集路径
-
out_dir:模型输出文件夹
-
batch_size：batch大小
其他超参数已经设置好。最后一个epoch结束
达到验收指标。
除了可以再训练过程中验证模型精度，可以使用stanet_eval_bone.py脚本进行测试，
动态模型为： 链接：https://pan.baidu.com/s/1LHsQ6nIUwoJVfHkDSIHHLA 提取码：y90l
!python ./STANET_Paddle/tutorials/eval/stanet_eval_bone.py --data_dir=./dataset/ --state_dict_path=./output1/stanet/best_model/model.pdparams参数介绍：
-
data_dir:数据集路径
-
weight_path:模型权重所在路径
输出如下：
2022-06-20 16:39:52 [INFO] Loading pretrained model from best_model/model.pdparams
2022-06-20 16:39:52 [INFO] There are 393/393 variables loaded into STANet.
2022-06-20 16:39:52 [INFO] Start to evaluate(total_samples=1024, total_steps=1024)...
OrderedDict([('miou', 0.9182049414147296), ('category_iou', array([0.99265375, 0.84375614])), ('oacc', 0.992934063076973), ('category_acc', array([0.99602765, 0.92133201])), ('kappa', 0.9115713464149582), ('category_F1-score', array([0.99631333, 0.91525785]))])可以将模型导出，动态图转静态图，使模型预测更快，可以使用stanet_export_bone..py脚本进行测试
在这里因为动静态模型转化的问题，修改了stanet的模型代码使其可以转出静态模型。
调试过程中参考这份文档 报错调试
!python ./STANET_Paddle/deploy/export/stanet_export_bone.py --state_dict_path=./output1/stanet/best_model/model.pdparams --save_dir=./inference_model_bone/ --fixed_input_shape=[1,3,256,256] 参数介绍：
- fixed_input_shape:预测图的形状 -save_dir 静态图导出路径
- state_dict_path:模型权重所在路径
可以使用stanet_infer.py脚本进行测试
!python ./STANET_Paddle/tutorials/infer/stanet_infer.py --infer_dir=./inference_model_bone --img_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset/ --output_dir=./STANET_Paddle/test_tipc/result/predict_output参数介绍：
- infer_dir:模型文件路径
- img_dir：用于推理的图片路径
- output_dir：预测结果输出路径
该部分依赖auto_log，需要进行安装，安装方式如下：
auto_log的详细介绍参考https://github.com/LDOUBLEV/AutoLog。
git clone https://github.com/LDOUBLEV/AutoLog
pip3 install -r requirements.txt
python3 setup.py bdist_wheel
pip3 install ./dist/auto_log-1.0.0-py3-none-any.whlbash ./STANET_Paddle/test_tipc/prepare.sh ./STANET_Paddle/test_tipc/configs/stanet/train_infer_python.txt 'lite_train_lite_infer'
bash ./STANET_Paddle/test_tipc/test_train_inference_python.sh ./STANET_Paddle/test_tipc/configs/stanet/train_infer_python.txt 'lite_train_lite_infer'测试结果如截图所示
aistudio@jupyter-2315405-4166493:~$ bash ./STANET_Paddle/test_tipc/test_train_inference_python.sh ./STANET_Paddle/test_tipc/configs/stanet/train_infer_python.txt 'lite_train_lite_infer'
[06-15 18:31:47 MainThread @logger.py:242] Argv: ./STANET_Paddle/tutorials/train/stanet_train_bonesmall.py --data_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --out_dir=./STANET_Paddle/test_tipc/result/stanet/ --num_epoch=6 --save_epoch=2 --batch_size=2
[06-15 18:31:47 MainThread @utils.py:79] WRN paddlepaddle version: 2.3.0. The dynamic graph version of PARL is under development, not fully tested and supported
2022-06-15 18:31:47,700-WARNING: type object 'QuantizationTransformPass' has no attribute '_supported_quantizable_op_type'
2022-06-15 18:31:47,701-WARNING: If you want to use training-aware and post-training quantization, please use Paddle >= 1.8.4 or develop version
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Models supported by PaddleClas
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
| Series | Name |
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
| AlexNet | AlexNet |
| DarkNet | DarkNet53 |
| DeiT | DeiT_base_distilled_patch16_224 DeiT_base_distilled_patch16_384 DeiT_base_patch16_224 DeiT_base_patch16_384 DeiT_small_distilled_patch16_224 |
| | DeiT_small_patch16_224 DeiT_tiny_distilled_patch16_224 DeiT_tiny_patch16_224 |
| DenseNet | DenseNet121 DenseNet161 DenseNet169 DenseNet201 DenseNet264 |
| DLA | DLA46_c DLA60x_c DLA34 DLA60 DLA60x DLA102 DLA102x DLA102x2 DLA169 |
| DPN | DPN68 DPN92 DPN98 DPN107 DPN131 |
| EfficientNet | EfficientNetB0 EfficientNetB0_small EfficientNetB1 EfficientNetB2 EfficientNetB3 EfficientNetB4 EfficientNetB5 EfficientNetB6 |
| | EfficientNetB7 |
| ESNet | ESNet_x0_25 ESNet_x0_5 ESNet_x0_75 ESNet_x1_0 |
| GhostNet | GhostNet_x0_5 GhostNet_x1_0 GhostNet_x1_3 GhostNet_x1_3_ssld |
| HarDNet | HarDNet39_ds HarDNet68_ds HarDNet68 HarDNet85 |
| HRNet | HRNet_W18_C HRNet_W30_C HRNet_W32_C HRNet_W40_C HRNet_W44_C HRNet_W48_C HRNet_W64_C HRNet_W18_C_ssld HRNet_W48_C_ssld |
| Inception | GoogLeNet InceptionV3 InceptionV4 |
| MixNet | MixNet_S MixNet_M MixNet_L |
| MobileNetV1 | MobileNetV1_x0_25 MobileNetV1_x0_5 MobileNetV1_x0_75 MobileNetV1 MobileNetV1_ssld |
| MobileNetV2 | MobileNetV2_x0_25 MobileNetV2_x0_5 MobileNetV2_x0_75 MobileNetV2 MobileNetV2_x1_5 MobileNetV2_x2_0 MobileNetV2_ssld |
| MobileNetV3 | MobileNetV3_small_x0_35 MobileNetV3_small_x0_5 MobileNetV3_small_x0_75 MobileNetV3_small_x1_0 MobileNetV3_small_x1_25 |
| | MobileNetV3_large_x0_35 MobileNetV3_large_x0_5 MobileNetV3_large_x0_75 MobileNetV3_large_x1_0 MobileNetV3_large_x1_25 |
| | MobileNetV3_small_x1_0_ssld MobileNetV3_large_x1_0_ssld |
| PPLCNet | PPLCNet_x0_25 PPLCNet_x0_35 PPLCNet_x0_5 PPLCNet_x0_75 PPLCNet_x1_0 PPLCNet_x1_5 PPLCNet_x2_0 PPLCNet_x2_5 |
| RedNet | RedNet26 RedNet38 RedNet50 RedNet101 RedNet152 |
| RegNet | RegNetX_4GF |
| Res2Net | Res2Net50_14w_8s Res2Net50_26w_4s Res2Net50_vd_26w_4s Res2Net200_vd_26w_4s Res2Net101_vd_26w_4s Res2Net50_vd_26w_4s_ssld |
| | Res2Net101_vd_26w_4s_ssld Res2Net200_vd_26w_4s_ssld |
| ResNeSt | ResNeSt50 ResNeSt50_fast_1s1x64d |
| ResNet | ResNet18 ResNet18_vd ResNet34 ResNet34_vd ResNet50 ResNet50_vc ResNet50_vd ResNet50_vd_v2 ResNet101 ResNet101_vd ResNet152 |
| | ResNet152_vd ResNet200_vd ResNet34_vd_ssld ResNet50_vd_ssld ResNet50_vd_ssld_v2 ResNet101_vd_ssld Fix_ResNet50_vd_ssld_v2 |
| | ResNet50_ACNet_deploy |
| ResNeXt | ResNeXt50_32x4d ResNeXt50_vd_32x4d ResNeXt50_64x4d ResNeXt50_vd_64x4d ResNeXt101_32x4d ResNeXt101_vd_32x4d ResNeXt101_32x8d_wsl |
| | ResNeXt101_32x16d_wsl ResNeXt101_32x32d_wsl ResNeXt101_32x48d_wsl Fix_ResNeXt101_32x48d_wsl ResNeXt101_64x4d ResNeXt101_vd_64x4d |
| | ResNeXt152_32x4d ResNeXt152_vd_32x4d ResNeXt152_64x4d ResNeXt152_vd_64x4d |
| ReXNet | ReXNet_1_0 ReXNet_1_3 ReXNet_1_5 ReXNet_2_0 ReXNet_3_0 |
| SENet | SENet154_vd SE_HRNet_W64_C_ssld SE_ResNet18_vd SE_ResNet34_vd SE_ResNet50_vd SE_ResNeXt50_32x4d SE_ResNeXt50_vd_32x4d SE_ResNeXt101_32x4d |
| ShuffleNetV2 | ShuffleNetV2_swish ShuffleNetV2_x0_25 ShuffleNetV2_x0_33 ShuffleNetV2_x0_5 ShuffleNetV2_x1_0 ShuffleNetV2_x1_5 ShuffleNetV2_x2_0 |
| SqueezeNet | SqueezeNet1_0 SqueezeNet1_1 |
| SwinTransformer | SwinTransformer_large_patch4_window7_224_22kto1k SwinTransformer_large_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window7_224_22kto1k SwinTransformer_base_patch4_window12_384_22kto1k SwinTransformer_base_patch4_window12_384 |
| | SwinTransformer_base_patch4_window7_224 SwinTransformer_small_patch4_window7_224 SwinTransformer_tiny_patch4_window7_224 |
| Twins | pcpvt_small pcpvt_base pcpvt_large alt_gvt_small alt_gvt_base alt_gvt_large |
| VGG | VGG11 VGG13 VGG16 VGG19 |
| VisionTransformer | ViT_base_patch16_224 ViT_base_patch16_384 ViT_base_patch32_384 ViT_large_patch16_224 ViT_large_patch16_384 ViT_large_patch32_384 |
| | ViT_small_patch16_224 |
| Xception | Xception41 Xception41_deeplab Xception65 Xception65_deeplab Xception71 |
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
Powered by PaddlePaddle!
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
2022-06-15 18:31:47 [INFO] 16 samples in file ./STANET_Paddle/test_tipc/data/mini_levir_dataset/train.txt
2022-06-15 18:31:47 [INFO] 8 samples in file ./STANET_Paddle/test_tipc/data/mini_levir_dataset/val.txt
W0615 18:31:47.809813 6982 gpu_context.cc:278] Please NOTE: device: 0, GPU Compute Capability: 8.0, Driver API Version: 11.2, Runtime API Version: 11.2
W0615 18:31:47.812942 6982 gpu_context.cc:306] device: 0, cuDNN Version: 8.2.
INFO 2022-06-15 18:31:50,207 logger.py:79] unique_endpoints {''}
[2022/06/15 18:31:50] root INFO: unique_endpoints {''}
INFO 2022-06-15 18:31:50,207 logger.py:79] Found /home/aistudio/.paddleclas/weights/ESNet_x1_0_pretrained.pdparams
[2022/06/15 18:31:50] root INFO: Found /home/aistudio/.paddleclas/weights/ESNet_x1_0_pretrained.pdparams
2022-06-15 18:31:55 [INFO] [TRAIN] Epoch 1 finished, loss=0.43874335 .
2022-06-15 18:31:56 [INFO] [TRAIN] Epoch 2 finished, loss=0.1586074 .
2022-06-15 18:31:56 [WARNING] Segmenter only supports batch_size=1 for each gpu/cpu card during evaluation, so batch_size is forcibly set to 1.
2022-06-15 18:31:56 [INFO] Start to evaluate(total_samples=8, total_steps=8)...
2022-06-15 18:31:57 [INFO] [EVAL] Finished, Epoch=2, miou=0.493079, category_iou=[0.98615837 0. ], oacc=0.986158, category_acc=[0.98615837 0. ], kappa=0.000000, category_F1-score=[0.99303095 0. ] .
2022-06-15 18:31:57 [INFO] Model saved in ./STANET_Paddle/test_tipc/result/stanet/best_model.
2022-06-15 18:31:57 [INFO] Current evaluated best model on eval_dataset is epoch_2, miou=0.49307918548583984
2022-06-15 18:31:57 [INFO] Model saved in ./STANET_Paddle/test_tipc/result/stanet/epoch_2.
2022-06-15 18:31:59 [INFO] [TRAIN] Epoch 3 finished, loss=0.15517747 .
2022-06-15 18:32:00 [INFO] [TRAIN] Epoch 4 finished, loss=0.13931209 .
2022-06-15 18:32:00 [WARNING] Segmenter only supports batch_size=1 for each gpu/cpu card during evaluation, so batch_size is forcibly set to 1.
2022-06-15 18:32:00 [INFO] Start to evaluate(total_samples=8, total_steps=8)...
2022-06-15 18:32:01 [INFO] [EVAL] Finished, Epoch=4, miou=0.493079, category_iou=[0.98615837 0. ], oacc=0.986158, category_acc=[0.98615837 0. ], kappa=0.000000, category_F1-score=[0.99303095 0. ] .
2022-06-15 18:32:01 [INFO] Current evaluated best model on eval_dataset is epoch_2, miou=0.49307918548583984
2022-06-15 18:32:01 [INFO] Model saved in ./STANET_Paddle/test_tipc/result/stanet/epoch_4.
2022-06-15 18:32:03 [INFO] [TRAIN] Epoch 5 finished, loss=0.17409432 .
2022-06-15 18:32:04 [INFO] [TRAIN] Epoch 6 finished, loss=0.13709755 .
2022-06-15 18:32:04 [WARNING] Segmenter only supports batch_size=1 for each gpu/cpu card during evaluation, so batch_size is forcibly set to 1.
2022-06-15 18:32:04 [INFO] Start to evaluate(total_samples=8, total_steps=8)...
2022-06-15 18:32:05 [INFO] [EVAL] Finished, Epoch=6, miou=0.492938, category_iou=[0.98587608 0. ], oacc=0.985876, category_acc=[0.98615446 0. ], kappa=-0.000554, category_F1-score=[0.99288782 nan] .
2022-06-15 18:32:05 [INFO] Current evaluated best model on eval_dataset is epoch_2, miou=0.49307918548583984
2022-06-15 18:32:05 [INFO] Model saved in ./STANET_Paddle/test_tipc/result/stanet/epoch_6.
Run successfully with command - python3.7 ./STANET_Paddle/tutorials/train/stanet_train_bonesmall.py --data_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --out_dir=./STANET_Paddle/test_tipc/result/stanet/ --num_epoch=6 --save_epoch=2 --batch_size=2 !
[06-15 18:32:12 MainThread @logger.py:242] Argv: ./STANET_Paddle/tutorials/eval/stanet_eval_bone.py --data_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --state_dict_path=./STANET_Paddle/test_tipc/result/stanet/best_model/model.pdparams
[06-15 18:32:12 MainThread @utils.py:79] WRN paddlepaddle version: 2.3.0. The dynamic graph version of PARL is under development, not fully tested and supported
2022-06-15 18:32:12,838-WARNING: type object 'QuantizationTransformPass' has no attribute '_supported_quantizable_op_type'
2022-06-15 18:32:12,838-WARNING: If you want to use training-aware and post-training quantization, please use Paddle >= 1.8.4 or develop version
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Models supported by PaddleClas
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
| Series | Name |
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
| AlexNet | AlexNet |
| DarkNet | DarkNet53 |
| DeiT | DeiT_base_distilled_patch16_224 DeiT_base_distilled_patch16_384 DeiT_base_patch16_224 DeiT_base_patch16_384 DeiT_small_distilled_patch16_224 |
| | DeiT_small_patch16_224 DeiT_tiny_distilled_patch16_224 DeiT_tiny_patch16_224 |
| DenseNet | DenseNet121 DenseNet161 DenseNet169 DenseNet201 DenseNet264 |
| DLA | DLA46_c DLA60x_c DLA34 DLA60 DLA60x DLA102 DLA102x DLA102x2 DLA169 |
| DPN | DPN68 DPN92 DPN98 DPN107 DPN131 |
| EfficientNet | EfficientNetB0 EfficientNetB0_small EfficientNetB1 EfficientNetB2 EfficientNetB3 EfficientNetB4 EfficientNetB5 EfficientNetB6 |
| | EfficientNetB7 |
| ESNet | ESNet_x0_25 ESNet_x0_5 ESNet_x0_75 ESNet_x1_0 |
| GhostNet | GhostNet_x0_5 GhostNet_x1_0 GhostNet_x1_3 GhostNet_x1_3_ssld |
| HarDNet | HarDNet39_ds HarDNet68_ds HarDNet68 HarDNet85 |
| HRNet | HRNet_W18_C HRNet_W30_C HRNet_W32_C HRNet_W40_C HRNet_W44_C HRNet_W48_C HRNet_W64_C HRNet_W18_C_ssld HRNet_W48_C_ssld |
| Inception | GoogLeNet InceptionV3 InceptionV4 |
| MixNet | MixNet_S MixNet_M MixNet_L |
| MobileNetV1 | MobileNetV1_x0_25 MobileNetV1_x0_5 MobileNetV1_x0_75 MobileNetV1 MobileNetV1_ssld |
| MobileNetV2 | MobileNetV2_x0_25 MobileNetV2_x0_5 MobileNetV2_x0_75 MobileNetV2 MobileNetV2_x1_5 MobileNetV2_x2_0 MobileNetV2_ssld |
| MobileNetV3 | MobileNetV3_small_x0_35 MobileNetV3_small_x0_5 MobileNetV3_small_x0_75 MobileNetV3_small_x1_0 MobileNetV3_small_x1_25 |
| | MobileNetV3_large_x0_35 MobileNetV3_large_x0_5 MobileNetV3_large_x0_75 MobileNetV3_large_x1_0 MobileNetV3_large_x1_25 |
| | MobileNetV3_small_x1_0_ssld MobileNetV3_large_x1_0_ssld |
| PPLCNet | PPLCNet_x0_25 PPLCNet_x0_35 PPLCNet_x0_5 PPLCNet_x0_75 PPLCNet_x1_0 PPLCNet_x1_5 PPLCNet_x2_0 PPLCNet_x2_5 |
| RedNet | RedNet26 RedNet38 RedNet50 RedNet101 RedNet152 |
| RegNet | RegNetX_4GF |
| Res2Net | Res2Net50_14w_8s Res2Net50_26w_4s Res2Net50_vd_26w_4s Res2Net200_vd_26w_4s Res2Net101_vd_26w_4s Res2Net50_vd_26w_4s_ssld |
| | Res2Net101_vd_26w_4s_ssld Res2Net200_vd_26w_4s_ssld |
| ResNeSt | ResNeSt50 ResNeSt50_fast_1s1x64d |
| ResNet | ResNet18 ResNet18_vd ResNet34 ResNet34_vd ResNet50 ResNet50_vc ResNet50_vd ResNet50_vd_v2 ResNet101 ResNet101_vd ResNet152 |
| | ResNet152_vd ResNet200_vd ResNet34_vd_ssld ResNet50_vd_ssld ResNet50_vd_ssld_v2 ResNet101_vd_ssld Fix_ResNet50_vd_ssld_v2 |
| | ResNet50_ACNet_deploy |
| ResNeXt | ResNeXt50_32x4d ResNeXt50_vd_32x4d ResNeXt50_64x4d ResNeXt50_vd_64x4d ResNeXt101_32x4d ResNeXt101_vd_32x4d ResNeXt101_32x8d_wsl |
| | ResNeXt101_32x16d_wsl ResNeXt101_32x32d_wsl ResNeXt101_32x48d_wsl Fix_ResNeXt101_32x48d_wsl ResNeXt101_64x4d ResNeXt101_vd_64x4d |
| | ResNeXt152_32x4d ResNeXt152_vd_32x4d ResNeXt152_64x4d ResNeXt152_vd_64x4d |
| ReXNet | ReXNet_1_0 ReXNet_1_3 ReXNet_1_5 ReXNet_2_0 ReXNet_3_0 |
| SENet | SENet154_vd SE_HRNet_W64_C_ssld SE_ResNet18_vd SE_ResNet34_vd SE_ResNet50_vd SE_ResNeXt50_32x4d SE_ResNeXt50_vd_32x4d SE_ResNeXt101_32x4d |
| ShuffleNetV2 | ShuffleNetV2_swish ShuffleNetV2_x0_25 ShuffleNetV2_x0_33 ShuffleNetV2_x0_5 ShuffleNetV2_x1_0 ShuffleNetV2_x1_5 ShuffleNetV2_x2_0 |
| SqueezeNet | SqueezeNet1_0 SqueezeNet1_1 |
| SwinTransformer | SwinTransformer_large_patch4_window7_224_22kto1k SwinTransformer_large_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window7_224_22kto1k SwinTransformer_base_patch4_window12_384_22kto1k SwinTransformer_base_patch4_window12_384 |
| | SwinTransformer_base_patch4_window7_224 SwinTransformer_small_patch4_window7_224 SwinTransformer_tiny_patch4_window7_224 |
| Twins | pcpvt_small pcpvt_base pcpvt_large alt_gvt_small alt_gvt_base alt_gvt_large |
| VGG | VGG11 VGG13 VGG16 VGG19 |
| VisionTransformer | ViT_base_patch16_224 ViT_base_patch16_384 ViT_base_patch32_384 ViT_large_patch16_224 ViT_large_patch16_384 ViT_large_patch32_384 |
| | ViT_small_patch16_224 |
| Xception | Xception41 Xception41_deeplab Xception65 Xception65_deeplab Xception71 |
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
Powered by PaddlePaddle!
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
2022-06-15 18:32:12 [INFO] 8 samples in file ./STANET_Paddle/test_tipc/data/mini_levir_dataset/val.txt
W0615 18:32:12.947710 7110 gpu_context.cc:278] Please NOTE: device: 0, GPU Compute Capability: 8.0, Driver API Version: 11.2, Runtime API Version: 11.2
W0615 18:32:12.950922 7110 gpu_context.cc:306] device: 0, cuDNN Version: 8.2.
INFO 2022-06-15 18:32:15,346 logger.py:79] unique_endpoints {''}
[2022/06/15 18:32:15] root INFO: unique_endpoints {''}
INFO 2022-06-15 18:32:15,346 logger.py:79] Found /home/aistudio/.paddleclas/weights/ESNet_x1_0_pretrained.pdparams
[2022/06/15 18:32:15] root INFO: Found /home/aistudio/.paddleclas/weights/ESNet_x1_0_pretrained.pdparams
2022-06-15 18:32:15 [INFO] Loading pretrained model from ./STANET_Paddle/test_tipc/result/stanet/best_model/model.pdparams
2022-06-15 18:32:15 [INFO] There are 393/393 variables loaded into STANet.
2022-06-15 18:32:15 [INFO] Start to evaluate(total_samples=8, total_steps=8)...
OrderedDict([('miou', 0.49307918548583984), ('category_iou', array([0.98615837, 0. ])), ('oacc', 0.9861583709716797), ('category_acc', array([0.98615837, 0. ])), ('kappa', 0.0), ('category_F1-score', array([0.99303095, 0. ]))])
Run successfully with command - python3.7 ./STANET_Paddle/tutorials/eval/stanet_eval_bone.py --data_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --state_dict_path=./STANET_Paddle/test_tipc/result/stanet/best_model/model.pdparams !
[06-15 18:32:24 MainThread @logger.py:242] Argv: ./STANET_Paddle/deploy/export/stanet_export_bone.py --state_dict_path=./STANET_Paddle/test_tipc/result/stanet/best_model/model.pdparams --save_dir=./STANET_Paddle/test_tipc/result/inference_model/ --fixed_input_shape=[1,3,256,256]
[06-15 18:32:24 MainThread @utils.py:79] WRN paddlepaddle version: 2.3.0. The dynamic graph version of PARL is under development, not fully tested and supported
2022-06-15 18:32:24,681-WARNING: type object 'QuantizationTransformPass' has no attribute '_supported_quantizable_op_type'
2022-06-15 18:32:24,681-WARNING: If you want to use training-aware and post-training quantization, please use Paddle >= 1.8.4 or develop version
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Models supported by PaddleClas
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
| Series | Name |
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
| AlexNet | AlexNet |
| DarkNet | DarkNet53 |
| DeiT | DeiT_base_distilled_patch16_224 DeiT_base_distilled_patch16_384 DeiT_base_patch16_224 DeiT_base_patch16_384 DeiT_small_distilled_patch16_224 |
| | DeiT_small_patch16_224 DeiT_tiny_distilled_patch16_224 DeiT_tiny_patch16_224 |
| DenseNet | DenseNet121 DenseNet161 DenseNet169 DenseNet201 DenseNet264 |
| DLA | DLA46_c DLA60x_c DLA34 DLA60 DLA60x DLA102 DLA102x DLA102x2 DLA169 |
| DPN | DPN68 DPN92 DPN98 DPN107 DPN131 |
| EfficientNet | EfficientNetB0 EfficientNetB0_small EfficientNetB1 EfficientNetB2 EfficientNetB3 EfficientNetB4 EfficientNetB5 EfficientNetB6 |
| | EfficientNetB7 |
| ESNet | ESNet_x0_25 ESNet_x0_5 ESNet_x0_75 ESNet_x1_0 |
| GhostNet | GhostNet_x0_5 GhostNet_x1_0 GhostNet_x1_3 GhostNet_x1_3_ssld |
| HarDNet | HarDNet39_ds HarDNet68_ds HarDNet68 HarDNet85 |
| HRNet | HRNet_W18_C HRNet_W30_C HRNet_W32_C HRNet_W40_C HRNet_W44_C HRNet_W48_C HRNet_W64_C HRNet_W18_C_ssld HRNet_W48_C_ssld |
| Inception | GoogLeNet InceptionV3 InceptionV4 |
| MixNet | MixNet_S MixNet_M MixNet_L |
| MobileNetV1 | MobileNetV1_x0_25 MobileNetV1_x0_5 MobileNetV1_x0_75 MobileNetV1 MobileNetV1_ssld |
| MobileNetV2 | MobileNetV2_x0_25 MobileNetV2_x0_5 MobileNetV2_x0_75 MobileNetV2 MobileNetV2_x1_5 MobileNetV2_x2_0 MobileNetV2_ssld |
| MobileNetV3 | MobileNetV3_small_x0_35 MobileNetV3_small_x0_5 MobileNetV3_small_x0_75 MobileNetV3_small_x1_0 MobileNetV3_small_x1_25 |
| | MobileNetV3_large_x0_35 MobileNetV3_large_x0_5 MobileNetV3_large_x0_75 MobileNetV3_large_x1_0 MobileNetV3_large_x1_25 |
| | MobileNetV3_small_x1_0_ssld MobileNetV3_large_x1_0_ssld |
| PPLCNet | PPLCNet_x0_25 PPLCNet_x0_35 PPLCNet_x0_5 PPLCNet_x0_75 PPLCNet_x1_0 PPLCNet_x1_5 PPLCNet_x2_0 PPLCNet_x2_5 |
| RedNet | RedNet26 RedNet38 RedNet50 RedNet101 RedNet152 |
| RegNet | RegNetX_4GF |
| Res2Net | Res2Net50_14w_8s Res2Net50_26w_4s Res2Net50_vd_26w_4s Res2Net200_vd_26w_4s Res2Net101_vd_26w_4s Res2Net50_vd_26w_4s_ssld |
| | Res2Net101_vd_26w_4s_ssld Res2Net200_vd_26w_4s_ssld |
| ResNeSt | ResNeSt50 ResNeSt50_fast_1s1x64d |
| ResNet | ResNet18 ResNet18_vd ResNet34 ResNet34_vd ResNet50 ResNet50_vc ResNet50_vd ResNet50_vd_v2 ResNet101 ResNet101_vd ResNet152 |
| | ResNet152_vd ResNet200_vd ResNet34_vd_ssld ResNet50_vd_ssld ResNet50_vd_ssld_v2 ResNet101_vd_ssld Fix_ResNet50_vd_ssld_v2 |
| | ResNet50_ACNet_deploy |
| ResNeXt | ResNeXt50_32x4d ResNeXt50_vd_32x4d ResNeXt50_64x4d ResNeXt50_vd_64x4d ResNeXt101_32x4d ResNeXt101_vd_32x4d ResNeXt101_32x8d_wsl |
| | ResNeXt101_32x16d_wsl ResNeXt101_32x32d_wsl ResNeXt101_32x48d_wsl Fix_ResNeXt101_32x48d_wsl ResNeXt101_64x4d ResNeXt101_vd_64x4d |
| | ResNeXt152_32x4d ResNeXt152_vd_32x4d ResNeXt152_64x4d ResNeXt152_vd_64x4d |
| ReXNet | ReXNet_1_0 ReXNet_1_3 ReXNet_1_5 ReXNet_2_0 ReXNet_3_0 |
| SENet | SENet154_vd SE_HRNet_W64_C_ssld SE_ResNet18_vd SE_ResNet34_vd SE_ResNet50_vd SE_ResNeXt50_32x4d SE_ResNeXt50_vd_32x4d SE_ResNeXt101_32x4d |
| ShuffleNetV2 | ShuffleNetV2_swish ShuffleNetV2_x0_25 ShuffleNetV2_x0_33 ShuffleNetV2_x0_5 ShuffleNetV2_x1_0 ShuffleNetV2_x1_5 ShuffleNetV2_x2_0 |
| SqueezeNet | SqueezeNet1_0 SqueezeNet1_1 |
| SwinTransformer | SwinTransformer_large_patch4_window7_224_22kto1k SwinTransformer_large_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window7_224_22kto1k SwinTransformer_base_patch4_window12_384_22kto1k SwinTransformer_base_patch4_window12_384 |
| | SwinTransformer_base_patch4_window7_224 SwinTransformer_small_patch4_window7_224 SwinTransformer_tiny_patch4_window7_224 |
| Twins | pcpvt_small pcpvt_base pcpvt_large alt_gvt_small alt_gvt_base alt_gvt_large |
| VGG | VGG11 VGG13 VGG16 VGG19 |
| VisionTransformer | ViT_base_patch16_224 ViT_base_patch16_384 ViT_base_patch32_384 ViT_large_patch16_224 ViT_large_patch16_384 ViT_large_patch32_384 |
| | ViT_small_patch16_224 |
| Xception | Xception41 Xception41_deeplab Xception65 Xception65_deeplab Xception71 |
+-------------------+--------------------------------------------------------------------------------------------------------------------------------------------------+
Powered by PaddlePaddle!
------------------------------------------------------------------------------------------------------------------------------------------------------------------------
W0615 18:32:24.788726 7290 gpu_context.cc:278] Please NOTE: device: 0, GPU Compute Capability: 8.0, Driver API Version: 11.2, Runtime API Version: 11.2
W0615 18:32:24.791882 7290 gpu_context.cc:306] device: 0, cuDNN Version: 8.2.
INFO 2022-06-15 18:32:27,175 logger.py:79] unique_endpoints {''}
[2022/06/15 18:32:27] root INFO: unique_endpoints {''}
INFO 2022-06-15 18:32:27,176 logger.py:79] Found /home/aistudio/.paddleclas/weights/ESNet_x1_0_pretrained.pdparams
[2022/06/15 18:32:27] root INFO: Found /home/aistudio/.paddleclas/weights/ESNet_x1_0_pretrained.pdparams
2022-06-15 18:32:27 [INFO] Loading pretrained model from ./STANET_Paddle/test_tipc/result/stanet/best_model/model.pdparams
2022-06-15 18:32:27 [INFO] There are 393/393 variables loaded into STANet.
2022-06-15 18:32:32 [INFO] The model for the inference deployment is saved in ./STANET_Paddle/test_tipc/result/inference_model/.
Run successfully with command - python3.7 ./STANET_Paddle/deploy/export/stanet_export_bone.py --state_dict_path=./STANET_Paddle/test_tipc/result/stanet/best_model/model.pdparams --save_dir=./STANET_Paddle/test_tipc/result/inference_model/ --fixed_input_shape=[1,3,256,256] !
[06-15 18:32:39 MainThread @logger.py:242] Argv: ./STANET_Paddle/tutorials/infer/stanet_infer.py --infer_dir=./STANET_Paddle/test_tipc/result/inference_model/ --img_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --output_dir=./STANET_Paddle/test_tipc/result/predict_output --use_gpu=True --precision=fp32 --enable_benchmark=True --use_tensorrt=False
[06-15 18:32:39 MainThread @utils.py:79] WRN paddlepaddle version: 2.3.0. The dynamic graph version of PARL is under development, not fully tested and supported
2022-06-15 18:32:39,611-WARNING: type object 'QuantizationTransformPass' has no attribute '_supported_quantizable_op_type'
2022-06-15 18:32:39,611-WARNING: If you want to use training-aware and post-training quantization, please use Paddle >= 1.8.4 or develop version
----------------------------------------------------------------------------------------------------------------------------
Models supported by PaddleClas
+-------------------+------------------------------------------------------------------------------------------------------+
| Series | Name |
+-------------------+------------------------------------------------------------------------------------------------------+
| AlexNet | AlexNet |
| DarkNet | DarkNet53 |
| DeiT | DeiT_base_distilled_patch16_224 DeiT_base_distilled_patch16_384 DeiT_base_patch16_224 |
| | DeiT_base_patch16_384 DeiT_small_distilled_patch16_224 DeiT_small_patch16_224 |
| | DeiT_tiny_distilled_patch16_224 DeiT_tiny_patch16_224 |
| DenseNet | DenseNet121 DenseNet161 DenseNet169 DenseNet201 DenseNet264 |
| DLA | DLA46_c DLA60x_c DLA34 DLA60 DLA60x DLA102 DLA102x DLA102x2 DLA169 |
| DPN | DPN68 DPN92 DPN98 DPN107 DPN131 |
| EfficientNet | EfficientNetB0 EfficientNetB0_small EfficientNetB1 EfficientNetB2 EfficientNetB3 EfficientNetB4 |
| | EfficientNetB5 EfficientNetB6 EfficientNetB7 |
| ESNet | ESNet_x0_25 ESNet_x0_5 ESNet_x0_75 ESNet_x1_0 |
| GhostNet | GhostNet_x0_5 GhostNet_x1_0 GhostNet_x1_3 GhostNet_x1_3_ssld |
| HarDNet | HarDNet39_ds HarDNet68_ds HarDNet68 HarDNet85 |
| HRNet | HRNet_W18_C HRNet_W30_C HRNet_W32_C HRNet_W40_C HRNet_W44_C HRNet_W48_C HRNet_W64_C |
| | HRNet_W18_C_ssld HRNet_W48_C_ssld |
| Inception | GoogLeNet InceptionV3 InceptionV4 |
| MixNet | MixNet_S MixNet_M MixNet_L |
| MobileNetV1 | MobileNetV1_x0_25 MobileNetV1_x0_5 MobileNetV1_x0_75 MobileNetV1 MobileNetV1_ssld |
| MobileNetV2 | MobileNetV2_x0_25 MobileNetV2_x0_5 MobileNetV2_x0_75 MobileNetV2 MobileNetV2_x1_5 |
| | MobileNetV2_x2_0 MobileNetV2_ssld |
| MobileNetV3 | MobileNetV3_small_x0_35 MobileNetV3_small_x0_5 MobileNetV3_small_x0_75 MobileNetV3_small_x1_0 |
| | MobileNetV3_small_x1_25 MobileNetV3_large_x0_35 MobileNetV3_large_x0_5 MobileNetV3_large_x0_75 |
| | MobileNetV3_large_x1_0 MobileNetV3_large_x1_25 MobileNetV3_small_x1_0_ssld |
| | MobileNetV3_large_x1_0_ssld |
| PPLCNet | PPLCNet_x0_25 PPLCNet_x0_35 PPLCNet_x0_5 PPLCNet_x0_75 PPLCNet_x1_0 PPLCNet_x1_5 PPLCNet_x2_0 |
| | PPLCNet_x2_5 |
| RedNet | RedNet26 RedNet38 RedNet50 RedNet101 RedNet152 |
| RegNet | RegNetX_4GF |
| Res2Net | Res2Net50_14w_8s Res2Net50_26w_4s Res2Net50_vd_26w_4s Res2Net200_vd_26w_4s Res2Net101_vd_26w_4s |
| | Res2Net50_vd_26w_4s_ssld Res2Net101_vd_26w_4s_ssld Res2Net200_vd_26w_4s_ssld |
| ResNeSt | ResNeSt50 ResNeSt50_fast_1s1x64d |
| ResNet | ResNet18 ResNet18_vd ResNet34 ResNet34_vd ResNet50 ResNet50_vc ResNet50_vd ResNet50_vd_v2 |
| | ResNet101 ResNet101_vd ResNet152 ResNet152_vd ResNet200_vd ResNet34_vd_ssld ResNet50_vd_ssld |
| | ResNet50_vd_ssld_v2 ResNet101_vd_ssld Fix_ResNet50_vd_ssld_v2 ResNet50_ACNet_deploy |
| ResNeXt | ResNeXt50_32x4d ResNeXt50_vd_32x4d ResNeXt50_64x4d ResNeXt50_vd_64x4d ResNeXt101_32x4d |
| | ResNeXt101_vd_32x4d ResNeXt101_32x8d_wsl ResNeXt101_32x16d_wsl ResNeXt101_32x32d_wsl |
| | ResNeXt101_32x48d_wsl Fix_ResNeXt101_32x48d_wsl ResNeXt101_64x4d ResNeXt101_vd_64x4d |
| | ResNeXt152_32x4d ResNeXt152_vd_32x4d ResNeXt152_64x4d ResNeXt152_vd_64x4d |
| ReXNet | ReXNet_1_0 ReXNet_1_3 ReXNet_1_5 ReXNet_2_0 ReXNet_3_0 |
| SENet | SENet154_vd SE_HRNet_W64_C_ssld SE_ResNet18_vd SE_ResNet34_vd SE_ResNet50_vd SE_ResNeXt50_32x4d |
| | SE_ResNeXt50_vd_32x4d SE_ResNeXt101_32x4d |
| ShuffleNetV2 | ShuffleNetV2_swish ShuffleNetV2_x0_25 ShuffleNetV2_x0_33 ShuffleNetV2_x0_5 ShuffleNetV2_x1_0 |
| | ShuffleNetV2_x1_5 ShuffleNetV2_x2_0 |
| SqueezeNet | SqueezeNet1_0 SqueezeNet1_1 |
| SwinTransformer | SwinTransformer_large_patch4_window7_224_22kto1k SwinTransformer_large_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window7_224_22kto1k SwinTransformer_base_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window12_384 SwinTransformer_base_patch4_window7_224 |
| | SwinTransformer_small_patch4_window7_224 SwinTransformer_tiny_patch4_window7_224 |
| Twins | pcpvt_small pcpvt_base pcpvt_large alt_gvt_small alt_gvt_base alt_gvt_large |
| VGG | VGG11 VGG13 VGG16 VGG19 |
| VisionTransformer | ViT_base_patch16_224 ViT_base_patch16_384 ViT_base_patch32_384 ViT_large_patch16_224 |
| | ViT_large_patch16_384 ViT_large_patch32_384 ViT_small_patch16_224 |
| Xception | Xception41 Xception41_deeplab Xception65 Xception65_deeplab Xception71 |
+-------------------+------------------------------------------------------------------------------------------------------+
Powered by PaddlePaddle!
----------------------------------------------------------------------------------------------------------------------------
2022-06-15 18:32:39 [INFO] Model[STANet] loaded.
total file number is 16
------------------ Inference Time Info ----------------------
total_time(ms): 3240.7, img_num: 1, batch_size: 1
average latency time(ms): 3240.70, QPS: 0.308575
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 3231.50, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 54.199999999999996, img_num: 1, batch_size: 1
average latency time(ms): 54.20, QPS: 18.450185
preprocess_time_per_im(ms): 8.80, inference_time_per_batch(ms): 45.20, postprocess_time_per_im(ms): 0.20
------------------ Inference Time Info ----------------------
total_time(ms): 58.9, img_num: 1, batch_size: 1
average latency time(ms): 58.90, QPS: 16.977929
preprocess_time_per_im(ms): 8.00, inference_time_per_batch(ms): 50.70, postprocess_time_per_im(ms): 0.20
------------------ Inference Time Info ----------------------
total_time(ms): 53.3, img_num: 1, batch_size: 1
average latency time(ms): 53.30, QPS: 18.761726
preprocess_time_per_im(ms): 8.40, inference_time_per_batch(ms): 44.80, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.3, img_num: 1, batch_size: 1
average latency time(ms): 52.30, QPS: 19.120459
preprocess_time_per_im(ms): 7.90, inference_time_per_batch(ms): 44.30, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.5, img_num: 1, batch_size: 1
average latency time(ms): 52.50, QPS: 19.047619
preprocess_time_per_im(ms): 8.10, inference_time_per_batch(ms): 44.30, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.3, img_num: 1, batch_size: 1
average latency time(ms): 52.30, QPS: 19.120459
preprocess_time_per_im(ms): 7.90, inference_time_per_batch(ms): 44.30, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.3, img_num: 1, batch_size: 1
average latency time(ms): 52.30, QPS: 19.120459
preprocess_time_per_im(ms): 8.00, inference_time_per_batch(ms): 44.20, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.2, img_num: 1, batch_size: 1
average latency time(ms): 52.20, QPS: 19.157088
preprocess_time_per_im(ms): 8.00, inference_time_per_batch(ms): 44.10, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 51.9, img_num: 1, batch_size: 1
average latency time(ms): 51.90, QPS: 19.267823
preprocess_time_per_im(ms): 7.70, inference_time_per_batch(ms): 44.10, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.3, img_num: 1, batch_size: 1
average latency time(ms): 52.30, QPS: 19.120459
preprocess_time_per_im(ms): 8.00, inference_time_per_batch(ms): 44.20, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.2, img_num: 1, batch_size: 1
average latency time(ms): 52.20, QPS: 19.157088
preprocess_time_per_im(ms): 7.80, inference_time_per_batch(ms): 44.20, postprocess_time_per_im(ms): 0.20
------------------ Inference Time Info ----------------------
total_time(ms): 52.3, img_num: 1, batch_size: 1
average latency time(ms): 52.30, QPS: 19.120459
preprocess_time_per_im(ms): 8.00, inference_time_per_batch(ms): 44.20, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.1, img_num: 1, batch_size: 1
average latency time(ms): 52.10, QPS: 19.193858
preprocess_time_per_im(ms): 8.00, inference_time_per_batch(ms): 44.00, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.1, img_num: 1, batch_size: 1
average latency time(ms): 52.10, QPS: 19.193858
preprocess_time_per_im(ms): 7.80, inference_time_per_batch(ms): 44.20, postprocess_time_per_im(ms): 0.10
------------------ Inference Time Info ----------------------
total_time(ms): 52.400000000000006, img_num: 1, batch_size: 1
average latency time(ms): 52.40, QPS: 19.083969
preprocess_time_per_im(ms): 8.00, inference_time_per_batch(ms): 44.20, postprocess_time_per_im(ms): 0.20
Run successfully with command - python3.7 ./STANET_Paddle/tutorials/infer/stanet_infer.py --infer_dir=./STANET_Paddle/test_tipc/result/inference_model/ --img_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --output_dir=./STANET_Paddle/test_tipc/result/predict_output --use_gpu=True --precision=fp32 --enable_benchmark=True --use_tensorrt=False > ./test_tipc/output/STANET/python_infer_gpu_usetrt_fp32_precision_null_batchsize_False.log 2>&1 !
[06-15 18:32:50 MainThread @logger.py:242] Argv: ./STANET_Paddle/tutorials/infer/stanet_infer.py --infer_dir=./STANET_Paddle/test_tipc/result/inference_model/ --img_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --output_dir=./STANET_Paddle/test_tipc/result/predict_output --use_gpu=False --enable_mkldnn=False --cpu_threads=1 --enable_benchmark=True --use_tensorrt=False
[06-15 18:32:50 MainThread @utils.py:79] WRN paddlepaddle version: 2.3.0. The dynamic graph version of PARL is under development, not fully tested and supported
2022-06-15 18:32:50,416-WARNING: type object 'QuantizationTransformPass' has no attribute '_supported_quantizable_op_type'
2022-06-15 18:32:50,416-WARNING: If you want to use training-aware and post-training quantization, please use Paddle >= 1.8.4 or develop version
----------------------------------------------------------------------------------------------------------------------------
Models supported by PaddleClas
+-------------------+------------------------------------------------------------------------------------------------------+
| Series | Name |
+-------------------+------------------------------------------------------------------------------------------------------+
| AlexNet | AlexNet |
| DarkNet | DarkNet53 |
| DeiT | DeiT_base_distilled_patch16_224 DeiT_base_distilled_patch16_384 DeiT_base_patch16_224 |
| | DeiT_base_patch16_384 DeiT_small_distilled_patch16_224 DeiT_small_patch16_224 |
| | DeiT_tiny_distilled_patch16_224 DeiT_tiny_patch16_224 |
| DenseNet | DenseNet121 DenseNet161 DenseNet169 DenseNet201 DenseNet264 |
| DLA | DLA46_c DLA60x_c DLA34 DLA60 DLA60x DLA102 DLA102x DLA102x2 DLA169 |
| DPN | DPN68 DPN92 DPN98 DPN107 DPN131 |
| EfficientNet | EfficientNetB0 EfficientNetB0_small EfficientNetB1 EfficientNetB2 EfficientNetB3 EfficientNetB4 |
| | EfficientNetB5 EfficientNetB6 EfficientNetB7 |
| ESNet | ESNet_x0_25 ESNet_x0_5 ESNet_x0_75 ESNet_x1_0 |
| GhostNet | GhostNet_x0_5 GhostNet_x1_0 GhostNet_x1_3 GhostNet_x1_3_ssld |
| HarDNet | HarDNet39_ds HarDNet68_ds HarDNet68 HarDNet85 |
| HRNet | HRNet_W18_C HRNet_W30_C HRNet_W32_C HRNet_W40_C HRNet_W44_C HRNet_W48_C HRNet_W64_C |
| | HRNet_W18_C_ssld HRNet_W48_C_ssld |
| Inception | GoogLeNet InceptionV3 InceptionV4 |
| MixNet | MixNet_S MixNet_M MixNet_L |
| MobileNetV1 | MobileNetV1_x0_25 MobileNetV1_x0_5 MobileNetV1_x0_75 MobileNetV1 MobileNetV1_ssld |
| MobileNetV2 | MobileNetV2_x0_25 MobileNetV2_x0_5 MobileNetV2_x0_75 MobileNetV2 MobileNetV2_x1_5 |
| | MobileNetV2_x2_0 MobileNetV2_ssld |
| MobileNetV3 | MobileNetV3_small_x0_35 MobileNetV3_small_x0_5 MobileNetV3_small_x0_75 MobileNetV3_small_x1_0 |
| | MobileNetV3_small_x1_25 MobileNetV3_large_x0_35 MobileNetV3_large_x0_5 MobileNetV3_large_x0_75 |
| | MobileNetV3_large_x1_0 MobileNetV3_large_x1_25 MobileNetV3_small_x1_0_ssld |
| | MobileNetV3_large_x1_0_ssld |
| PPLCNet | PPLCNet_x0_25 PPLCNet_x0_35 PPLCNet_x0_5 PPLCNet_x0_75 PPLCNet_x1_0 PPLCNet_x1_5 PPLCNet_x2_0 |
| | PPLCNet_x2_5 |
| RedNet | RedNet26 RedNet38 RedNet50 RedNet101 RedNet152 |
| RegNet | RegNetX_4GF |
| Res2Net | Res2Net50_14w_8s Res2Net50_26w_4s Res2Net50_vd_26w_4s Res2Net200_vd_26w_4s Res2Net101_vd_26w_4s |
| | Res2Net50_vd_26w_4s_ssld Res2Net101_vd_26w_4s_ssld Res2Net200_vd_26w_4s_ssld |
| ResNeSt | ResNeSt50 ResNeSt50_fast_1s1x64d |
| ResNet | ResNet18 ResNet18_vd ResNet34 ResNet34_vd ResNet50 ResNet50_vc ResNet50_vd ResNet50_vd_v2 |
| | ResNet101 ResNet101_vd ResNet152 ResNet152_vd ResNet200_vd ResNet34_vd_ssld ResNet50_vd_ssld |
| | ResNet50_vd_ssld_v2 ResNet101_vd_ssld Fix_ResNet50_vd_ssld_v2 ResNet50_ACNet_deploy |
| ResNeXt | ResNeXt50_32x4d ResNeXt50_vd_32x4d ResNeXt50_64x4d ResNeXt50_vd_64x4d ResNeXt101_32x4d |
| | ResNeXt101_vd_32x4d ResNeXt101_32x8d_wsl ResNeXt101_32x16d_wsl ResNeXt101_32x32d_wsl |
| | ResNeXt101_32x48d_wsl Fix_ResNeXt101_32x48d_wsl ResNeXt101_64x4d ResNeXt101_vd_64x4d |
| | ResNeXt152_32x4d ResNeXt152_vd_32x4d ResNeXt152_64x4d ResNeXt152_vd_64x4d |
| ReXNet | ReXNet_1_0 ReXNet_1_3 ReXNet_1_5 ReXNet_2_0 ReXNet_3_0 |
| SENet | SENet154_vd SE_HRNet_W64_C_ssld SE_ResNet18_vd SE_ResNet34_vd SE_ResNet50_vd SE_ResNeXt50_32x4d |
| | SE_ResNeXt50_vd_32x4d SE_ResNeXt101_32x4d |
| ShuffleNetV2 | ShuffleNetV2_swish ShuffleNetV2_x0_25 ShuffleNetV2_x0_33 ShuffleNetV2_x0_5 ShuffleNetV2_x1_0 |
| | ShuffleNetV2_x1_5 ShuffleNetV2_x2_0 |
| SqueezeNet | SqueezeNet1_0 SqueezeNet1_1 |
| SwinTransformer | SwinTransformer_large_patch4_window7_224_22kto1k SwinTransformer_large_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window7_224_22kto1k SwinTransformer_base_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window12_384 SwinTransformer_base_patch4_window7_224 |
| | SwinTransformer_small_patch4_window7_224 SwinTransformer_tiny_patch4_window7_224 |
| Twins | pcpvt_small pcpvt_base pcpvt_large alt_gvt_small alt_gvt_base alt_gvt_large |
| VGG | VGG11 VGG13 VGG16 VGG19 |
| VisionTransformer | ViT_base_patch16_224 ViT_base_patch16_384 ViT_base_patch32_384 ViT_large_patch16_224 |
| | ViT_large_patch16_384 ViT_large_patch32_384 ViT_small_patch16_224 |
| Xception | Xception41 Xception41_deeplab Xception65 Xception65_deeplab Xception71 |
+-------------------+------------------------------------------------------------------------------------------------------+
Powered by PaddlePaddle!
----------------------------------------------------------------------------------------------------------------------------
2022-06-15 18:32:50 [INFO] Model[STANet] loaded.
--- fused 0 elementwise_add with relu activation
--- fused 0 elementwise_add with tanh activation
--- fused 0 elementwise_add with leaky_relu activation
--- fused 0 elementwise_add with swish activation
--- fused 0 elementwise_add with hardswish activation
--- fused 0 elementwise_add with sqrt activation
--- fused 0 elementwise_add with abs activation
--- fused 0 elementwise_add with clip activation
--- fused 0 elementwise_add with gelu activation
--- fused 0 elementwise_add with relu6 activation
--- fused 0 elementwise_add with sigmoid activation
--- fused 0 elementwise_sub with relu activation
--- fused 0 elementwise_sub with tanh activation
--- fused 0 elementwise_sub with leaky_relu activation
--- fused 0 elementwise_sub with swish activation
--- fused 0 elementwise_sub with hardswish activation
--- fused 0 elementwise_sub with sqrt activation
--- fused 1 elementwise_sub with abs activation
--- fused 0 elementwise_sub with clip activation
--- fused 0 elementwise_sub with gelu activation
--- fused 0 elementwise_sub with relu6 activation
--- fused 0 elementwise_sub with sigmoid activation
--- fused 0 elementwise_mul with relu activation
--- fused 0 elementwise_mul with tanh activation
--- fused 0 elementwise_mul with leaky_relu activation
--- fused 0 elementwise_mul with swish activation
--- fused 0 elementwise_mul with hardswish activation
--- fused 0 elementwise_mul with sqrt activation
--- fused 0 elementwise_mul with abs activation
--- fused 0 elementwise_mul with clip activation
--- fused 0 elementwise_mul with gelu activation
--- fused 0 elementwise_mul with relu6 activation
--- fused 0 elementwise_mul with sigmoid activation
total file number is 16
------------------ Inference Time Info ----------------------
total_time(ms): 981.1, img_num: 1, batch_size: 1
average latency time(ms): 981.10, QPS: 1.019264
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 971.70, postprocess_time_per_im(ms): 0.50
------------------ Inference Time Info ----------------------
total_time(ms): 677.2, img_num: 1, batch_size: 1
average latency time(ms): 677.20, QPS: 1.476669
preprocess_time_per_im(ms): 9.00, inference_time_per_batch(ms): 667.70, postprocess_time_per_im(ms): 0.50
------------------ Inference Time Info ----------------------
total_time(ms): 1377.0, img_num: 1, batch_size: 1
average latency time(ms): 1377.00, QPS: 0.726216
preprocess_time_per_im(ms): 9.40, inference_time_per_batch(ms): 1367.20, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 1015.9, img_num: 1, batch_size: 1
average latency time(ms): 1015.90, QPS: 0.984349
preprocess_time_per_im(ms): 9.40, inference_time_per_batch(ms): 1006.20, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 756.1, img_num: 1, batch_size: 1
average latency time(ms): 756.10, QPS: 1.322576
preprocess_time_per_im(ms): 9.00, inference_time_per_batch(ms): 746.70, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 747.4, img_num: 1, batch_size: 1
average latency time(ms): 747.40, QPS: 1.337972
preprocess_time_per_im(ms): 9.00, inference_time_per_batch(ms): 738.00, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 876.3, img_num: 1, batch_size: 1
average latency time(ms): 876.30, QPS: 1.141162
preprocess_time_per_im(ms): 9.10, inference_time_per_batch(ms): 866.70, postprocess_time_per_im(ms): 0.50
------------------ Inference Time Info ----------------------
total_time(ms): 781.2, img_num: 1, batch_size: 1
average latency time(ms): 781.20, QPS: 1.280082
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 772.00, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 1018.4, img_num: 1, batch_size: 1
average latency time(ms): 1018.40, QPS: 0.981932
preprocess_time_per_im(ms): 8.70, inference_time_per_batch(ms): 1009.30, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 1153.0, img_num: 1, batch_size: 1
average latency time(ms): 1153.00, QPS: 0.867303
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 1143.70, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 663.8, img_num: 1, batch_size: 1
average latency time(ms): 663.80, QPS: 1.506478
preprocess_time_per_im(ms): 9.10, inference_time_per_batch(ms): 654.40, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 687.2, img_num: 1, batch_size: 1
average latency time(ms): 687.20, QPS: 1.455180
preprocess_time_per_im(ms): 8.40, inference_time_per_batch(ms): 678.40, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 837.3000000000001, img_num: 1, batch_size: 1
average latency time(ms): 837.30, QPS: 1.194315
preprocess_time_per_im(ms): 9.30, inference_time_per_batch(ms): 827.60, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 803.2, img_num: 1, batch_size: 1
average latency time(ms): 803.20, QPS: 1.245020
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 793.90, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 781.8000000000001, img_num: 1, batch_size: 1
average latency time(ms): 781.80, QPS: 1.279100
preprocess_time_per_im(ms): 9.10, inference_time_per_batch(ms): 772.40, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 828.3000000000001, img_num: 1, batch_size: 1
average latency time(ms): 828.30, QPS: 1.207292
preprocess_time_per_im(ms): 8.70, inference_time_per_batch(ms): 819.30, postprocess_time_per_im(ms): 0.30
Run successfully with command - python3.7 ./STANET_Paddle/tutorials/infer/stanet_infer.py --infer_dir=./STANET_Paddle/test_tipc/result/inference_model/ --img_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --output_dir=./STANET_Paddle/test_tipc/result/predict_output --use_gpu=False --enable_mkldnn=False --cpu_threads=1 --enable_benchmark=True --use_tensorrt=False > ./test_tipc/output/STANET/python_infer_cpu_usemkldnn_False_threads_1_precision_null_batchsize_False.log 2>&1 !
[06-15 18:33:10 MainThread @logger.py:242] Argv: ./STANET_Paddle/tutorials/infer/stanet_infer.py --infer_dir=./STANET_Paddle/test_tipc/result/inference_model/ --img_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --output_dir=./STANET_Paddle/test_tipc/result/predict_output --use_gpu=False --enable_mkldnn=False --cpu_threads=2 --enable_benchmark=True --use_tensorrt=False
[06-15 18:33:10 MainThread @utils.py:79] WRN paddlepaddle version: 2.3.0. The dynamic graph version of PARL is under development, not fully tested and supported
2022-06-15 18:33:10,834-WARNING: type object 'QuantizationTransformPass' has no attribute '_supported_quantizable_op_type'
2022-06-15 18:33:10,834-WARNING: If you want to use training-aware and post-training quantization, please use Paddle >= 1.8.4 or develop version
----------------------------------------------------------------------------------------------------------------------------
Models supported by PaddleClas
+-------------------+------------------------------------------------------------------------------------------------------+
| Series | Name |
+-------------------+------------------------------------------------------------------------------------------------------+
| AlexNet | AlexNet |
| DarkNet | DarkNet53 |
| DeiT | DeiT_base_distilled_patch16_224 DeiT_base_distilled_patch16_384 DeiT_base_patch16_224 |
| | DeiT_base_patch16_384 DeiT_small_distilled_patch16_224 DeiT_small_patch16_224 |
| | DeiT_tiny_distilled_patch16_224 DeiT_tiny_patch16_224 |
| DenseNet | DenseNet121 DenseNet161 DenseNet169 DenseNet201 DenseNet264 |
| DLA | DLA46_c DLA60x_c DLA34 DLA60 DLA60x DLA102 DLA102x DLA102x2 DLA169 |
| DPN | DPN68 DPN92 DPN98 DPN107 DPN131 |
| EfficientNet | EfficientNetB0 EfficientNetB0_small EfficientNetB1 EfficientNetB2 EfficientNetB3 EfficientNetB4 |
| | EfficientNetB5 EfficientNetB6 EfficientNetB7 |
| ESNet | ESNet_x0_25 ESNet_x0_5 ESNet_x0_75 ESNet_x1_0 |
| GhostNet | GhostNet_x0_5 GhostNet_x1_0 GhostNet_x1_3 GhostNet_x1_3_ssld |
| HarDNet | HarDNet39_ds HarDNet68_ds HarDNet68 HarDNet85 |
| HRNet | HRNet_W18_C HRNet_W30_C HRNet_W32_C HRNet_W40_C HRNet_W44_C HRNet_W48_C HRNet_W64_C |
| | HRNet_W18_C_ssld HRNet_W48_C_ssld |
| Inception | GoogLeNet InceptionV3 InceptionV4 |
| MixNet | MixNet_S MixNet_M MixNet_L |
| MobileNetV1 | MobileNetV1_x0_25 MobileNetV1_x0_5 MobileNetV1_x0_75 MobileNetV1 MobileNetV1_ssld |
| MobileNetV2 | MobileNetV2_x0_25 MobileNetV2_x0_5 MobileNetV2_x0_75 MobileNetV2 MobileNetV2_x1_5 |
| | MobileNetV2_x2_0 MobileNetV2_ssld |
| MobileNetV3 | MobileNetV3_small_x0_35 MobileNetV3_small_x0_5 MobileNetV3_small_x0_75 MobileNetV3_small_x1_0 |
| | MobileNetV3_small_x1_25 MobileNetV3_large_x0_35 MobileNetV3_large_x0_5 MobileNetV3_large_x0_75 |
| | MobileNetV3_large_x1_0 MobileNetV3_large_x1_25 MobileNetV3_small_x1_0_ssld |
| | MobileNetV3_large_x1_0_ssld |
| PPLCNet | PPLCNet_x0_25 PPLCNet_x0_35 PPLCNet_x0_5 PPLCNet_x0_75 PPLCNet_x1_0 PPLCNet_x1_5 PPLCNet_x2_0 |
| | PPLCNet_x2_5 |
| RedNet | RedNet26 RedNet38 RedNet50 RedNet101 RedNet152 |
| RegNet | RegNetX_4GF |
| Res2Net | Res2Net50_14w_8s Res2Net50_26w_4s Res2Net50_vd_26w_4s Res2Net200_vd_26w_4s Res2Net101_vd_26w_4s |
| | Res2Net50_vd_26w_4s_ssld Res2Net101_vd_26w_4s_ssld Res2Net200_vd_26w_4s_ssld |
| ResNeSt | ResNeSt50 ResNeSt50_fast_1s1x64d |
| ResNet | ResNet18 ResNet18_vd ResNet34 ResNet34_vd ResNet50 ResNet50_vc ResNet50_vd ResNet50_vd_v2 |
| | ResNet101 ResNet101_vd ResNet152 ResNet152_vd ResNet200_vd ResNet34_vd_ssld ResNet50_vd_ssld |
| | ResNet50_vd_ssld_v2 ResNet101_vd_ssld Fix_ResNet50_vd_ssld_v2 ResNet50_ACNet_deploy |
| ResNeXt | ResNeXt50_32x4d ResNeXt50_vd_32x4d ResNeXt50_64x4d ResNeXt50_vd_64x4d ResNeXt101_32x4d |
| | ResNeXt101_vd_32x4d ResNeXt101_32x8d_wsl ResNeXt101_32x16d_wsl ResNeXt101_32x32d_wsl |
| | ResNeXt101_32x48d_wsl Fix_ResNeXt101_32x48d_wsl ResNeXt101_64x4d ResNeXt101_vd_64x4d |
| | ResNeXt152_32x4d ResNeXt152_vd_32x4d ResNeXt152_64x4d ResNeXt152_vd_64x4d |
| ReXNet | ReXNet_1_0 ReXNet_1_3 ReXNet_1_5 ReXNet_2_0 ReXNet_3_0 |
| SENet | SENet154_vd SE_HRNet_W64_C_ssld SE_ResNet18_vd SE_ResNet34_vd SE_ResNet50_vd SE_ResNeXt50_32x4d |
| | SE_ResNeXt50_vd_32x4d SE_ResNeXt101_32x4d |
| ShuffleNetV2 | ShuffleNetV2_swish ShuffleNetV2_x0_25 ShuffleNetV2_x0_33 ShuffleNetV2_x0_5 ShuffleNetV2_x1_0 |
| | ShuffleNetV2_x1_5 ShuffleNetV2_x2_0 |
| SqueezeNet | SqueezeNet1_0 SqueezeNet1_1 |
| SwinTransformer | SwinTransformer_large_patch4_window7_224_22kto1k SwinTransformer_large_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window7_224_22kto1k SwinTransformer_base_patch4_window12_384_22kto1k |
| | SwinTransformer_base_patch4_window12_384 SwinTransformer_base_patch4_window7_224 |
| | SwinTransformer_small_patch4_window7_224 SwinTransformer_tiny_patch4_window7_224 |
| Twins | pcpvt_small pcpvt_base pcpvt_large alt_gvt_small alt_gvt_base alt_gvt_large |
| VGG | VGG11 VGG13 VGG16 VGG19 |
| VisionTransformer | ViT_base_patch16_224 ViT_base_patch16_384 ViT_base_patch32_384 ViT_large_patch16_224 |
| | ViT_large_patch16_384 ViT_large_patch32_384 ViT_small_patch16_224 |
| Xception | Xception41 Xception41_deeplab Xception65 Xception65_deeplab Xception71 |
+-------------------+------------------------------------------------------------------------------------------------------+
Powered by PaddlePaddle!
----------------------------------------------------------------------------------------------------------------------------
2022-06-15 18:33:10 [INFO] Model[STANet] loaded.
--- fused 0 elementwise_add with relu activation
--- fused 0 elementwise_add with tanh activation
--- fused 0 elementwise_add with leaky_relu activation
--- fused 0 elementwise_add with swish activation
--- fused 0 elementwise_add with hardswish activation
--- fused 0 elementwise_add with sqrt activation
--- fused 0 elementwise_add with abs activation
--- fused 0 elementwise_add with clip activation
--- fused 0 elementwise_add with gelu activation
--- fused 0 elementwise_add with relu6 activation
--- fused 0 elementwise_add with sigmoid activation
--- fused 0 elementwise_sub with relu activation
--- fused 0 elementwise_sub with tanh activation
--- fused 0 elementwise_sub with leaky_relu activation
--- fused 0 elementwise_sub with swish activation
--- fused 0 elementwise_sub with hardswish activation
--- fused 0 elementwise_sub with sqrt activation
--- fused 1 elementwise_sub with abs activation
--- fused 0 elementwise_sub with clip activation
--- fused 0 elementwise_sub with gelu activation
--- fused 0 elementwise_sub with relu6 activation
--- fused 0 elementwise_sub with sigmoid activation
--- fused 0 elementwise_mul with relu activation
--- fused 0 elementwise_mul with tanh activation
--- fused 0 elementwise_mul with leaky_relu activation
--- fused 0 elementwise_mul with swish activation
--- fused 0 elementwise_mul with hardswish activation
--- fused 0 elementwise_mul with sqrt activation
--- fused 0 elementwise_mul with abs activation
--- fused 0 elementwise_mul with clip activation
--- fused 0 elementwise_mul with gelu activation
--- fused 0 elementwise_mul with relu6 activation
--- fused 0 elementwise_mul with sigmoid activation
total file number is 16
------------------ Inference Time Info ----------------------
total_time(ms): 1012.8, img_num: 1, batch_size: 1
average latency time(ms): 1012.80, QPS: 0.987362
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 1003.40, postprocess_time_per_im(ms): 0.50
------------------ Inference Time Info ----------------------
total_time(ms): 678.9, img_num: 1, batch_size: 1
average latency time(ms): 678.90, QPS: 1.472971
preprocess_time_per_im(ms): 9.20, inference_time_per_batch(ms): 669.40, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 885.6, img_num: 1, batch_size: 1
average latency time(ms): 885.60, QPS: 1.129178
preprocess_time_per_im(ms): 8.80, inference_time_per_batch(ms): 876.50, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 802.0, img_num: 1, batch_size: 1
average latency time(ms): 802.00, QPS: 1.246883
preprocess_time_per_im(ms): 9.30, inference_time_per_batch(ms): 792.10, postprocess_time_per_im(ms): 0.60
------------------ Inference Time Info ----------------------
total_time(ms): 669.1, img_num: 1, batch_size: 1
average latency time(ms): 669.10, QPS: 1.494545
preprocess_time_per_im(ms): 9.70, inference_time_per_batch(ms): 659.10, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 671.4, img_num: 1, batch_size: 1
average latency time(ms): 671.40, QPS: 1.489425
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 662.20, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 687.0, img_num: 1, batch_size: 1
average latency time(ms): 687.00, QPS: 1.455604
preprocess_time_per_im(ms): 8.80, inference_time_per_batch(ms): 677.70, postprocess_time_per_im(ms): 0.50
------------------ Inference Time Info ----------------------
total_time(ms): 845.7, img_num: 1, batch_size: 1
average latency time(ms): 845.70, QPS: 1.182452
preprocess_time_per_im(ms): 9.70, inference_time_per_batch(ms): 835.70, postprocess_time_per_im(ms): 0.30
------------------ Inference Time Info ----------------------
total_time(ms): 794.5, img_num: 1, batch_size: 1
average latency time(ms): 794.50, QPS: 1.258653
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 785.20, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 951.6, img_num: 1, batch_size: 1
average latency time(ms): 951.60, QPS: 1.050862
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 942.20, postprocess_time_per_im(ms): 0.50
------------------ Inference Time Info ----------------------
total_time(ms): 930.0, img_num: 1, batch_size: 1
average latency time(ms): 930.00, QPS: 1.075269
preprocess_time_per_im(ms): 9.50, inference_time_per_batch(ms): 920.10, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 1092.5, img_num: 1, batch_size: 1
average latency time(ms): 1092.50, QPS: 0.915332
preprocess_time_per_im(ms): 8.80, inference_time_per_batch(ms): 1083.20, postprocess_time_per_im(ms): 0.50
------------------ Inference Time Info ----------------------
total_time(ms): 949.6, img_num: 1, batch_size: 1
average latency time(ms): 949.60, QPS: 1.053075
preprocess_time_per_im(ms): 9.30, inference_time_per_batch(ms): 939.60, postprocess_time_per_im(ms): 0.70
------------------ Inference Time Info ----------------------
total_time(ms): 996.0, img_num: 1, batch_size: 1
average latency time(ms): 996.00, QPS: 1.004016
preprocess_time_per_im(ms): 9.60, inference_time_per_batch(ms): 986.00, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 948.4, img_num: 1, batch_size: 1
average latency time(ms): 948.40, QPS: 1.054407
preprocess_time_per_im(ms): 9.60, inference_time_per_batch(ms): 938.40, postprocess_time_per_im(ms): 0.40
------------------ Inference Time Info ----------------------
total_time(ms): 820.6, img_num: 1, batch_size: 1
average latency time(ms): 820.60, QPS: 1.218621
preprocess_time_per_im(ms): 8.90, inference_time_per_batch(ms): 811.30, postprocess_time_per_im(ms): 0.40
Run successfully with command - python3.7 ./STANET_Paddle/tutorials/infer/stanet_infer.py --infer_dir=./STANET_Paddle/test_tipc/result/inference_model/ --img_dir=./STANET_Paddle/test_tipc/data/mini_levir_dataset --output_dir=./STANET_Paddle/test_tipc/result/predict_output --use_gpu=False --enable_mkldnn=False --cpu_threads=2 --enable_benchmark=True --use_tensorrt=False > ./test_tipc/output/STANET/python_infer_cpu_usemkldnn_False_threads_2_precision_null_batchsize_False.log 2>&1 !
aistudio@jupyter-2315405-4166493:~$
StaNet-Paddle
├── deploy # 部署相关的文档和脚本
├── docs # 整个项目图片
├── output # 输出的VDL日志
├── test_tipc # tipc程序相关
├── paddlers
│ ├── custom_models # 自定义网络模型代码
│ ├── datasets # 数据加载相关代码
│ ├── models # 套件网络模型代码
│ ├── tasks # 相关任务代码
│ ├── tools # 相关脚本
│ ├── transforms # 数据处理及增强相关代码
│ └── utils # 各种实用程序文件
├── tools # 用于处理遥感数据的脚本
└── tutorials
└── train # 模型训练
└── eval # 模型评估和TIPC训练
└── infer # 模型推理
└── predict # 动态模型预测
| 信息 | 描述 |
|---|---|
| 模型名称 | STANET |
| 框架版本 | PaddlePaddle==2.2.0 |
| 应用场景 | 遥感图像变化检测 |
感谢百度提供的算力，以及举办的本场比赛，让我增强对paddle的熟练度，加深对变化检测模型的理解，加深对模型压缩的理解！
部分功能参考 PaddleRS 手把手教你PaddleRS实现变化检测
部分功能参考 基于Paddle复现SNUNet-CD，
首先，这篇文章对我，做模型挑战的帮助很大，给了我很多思路《产业级SOTA模型优化指南》，我按照指南一步一步走。
esnet在imagenet的精度为ImageNet/Acc 0.7392，原模型使用的backbone为res18精度为ImageNet/Acc 0.7098，backbone的特征提取能力增强，模型的大小也大大降低，所以我选它作为backbone。我这次只用到了其中的cov1和blocks两个组件。其他组件例如maxpool，我认为可能对基于语义分割的任务意义不大，所以我去掉了（注意要将模型中用不到的部分去掉，来减少参数量）。
模型训练很快就出现过拟合现象，我选择使用多种数据增广方式来缓解 。例如RandomHorizontalFlip、RandomVerticalFlip、MixupImage、RandomDistort、RandomBlur、RandomSwap。
原模型使用backbone的方式，是将backbone的多尺寸输出的c统一降低到90，之后尺寸放大到相同的大小，这带来了 一定的精度损失，我修改这一点通过1*1卷积根据输出的维度，处理后仍为相同的维度。
知识蒸馏我认为是种很好的降低模型大小的方案，原模型因为模型大小还未发挥全部的实力，增大模型的参数，后知识蒸馏可能会带来不错的效果。