The ResNext models trained with billion scale weakly-supervised data. Please refer to Exploring the Limits of Weakly Supervised Pretraining presented at ECCV 2018 for the details of model training. The Caffe models are converted from PyTorch.
| Network | Top-1 Acc. | Top-5 Acc. | Google Drive | Baidu Yun |
|---|---|---|---|---|
| ResNeXt-101 32x8d | 82.2 | 96.4 | link |
link didj |
| ResNeXt-101 32x16d | 84.2 | 97.2 | link |
link wtru |
| ResNeXt-101 32x32d | 85.1 | 97.5 | link |
link jc9e |
Note: Testing Input Size: 224 x 224
Evaluate ResNeXt-101 32x8d:
python2 eval_img.py --proto resnext-101-32x8d-wsl.prototxt --model resnext-101-32x8d-wsl.caffemodel --image imgs/cat.jpg
Expected Outputs:
0.82 - 'n02124075 Egyptian cat'
0.09 - 'n02123045 tabby, tabby cat'
0.04 - 'n02127052 lynx, catamount'
0.03 - 'n02123159 tiger cat'
0.00 - 'n02123394 Persian cat'Evaluate ResNeXt-101 32x16d:
python2 eval_img.py --proto resnext-101-32x16d-wsl.prototxt --model resnext-101-32x16d-wsl.caffemodel --image imgs/cat.jpg
Expected Outputs:
0.85 - 'n02124075 Egyptian cat'
0.06 - 'n02127052 lynx, catamount'
0.04 - 'n02123159 tiger cat'
0.03 - 'n02123045 tabby, tabby cat'
0.00 - 'n02123394 Persian cat'Evaluate ResNeXt-101 32x32d:
python2 eval_img.py --proto resnext-101-32x32d-wsl.prototxt --model resnext-101-32x32d-wsl.caffemodel --image imgs/cat.jpg
Expected Outputs:
0.91 - 'n02124075 Egyptian cat'
0.04 - 'n02123045 tabby, tabby cat'
0.02 - 'n02127052 lynx, catamount'
0.01 - 'n02123159 tiger cat'
0.00 - 'n02123394 Persian cat'python3 converter/pt_to_caffe.py --save-root tmp --net resnext-101-32x48d
Requirements:
- PyTorch >= 1.0.1
- Python >= 3.5
- torchvision >= 0.2.0
A problem will be encountered when converting the largest model ResNeXt-101 32x48d. If you have enough space to make it work, welcome to contribute.
Transform Completed
[libprotobuf FATAL google/protobuf/wire_format.cc:752] CHECK failed: (output->ByteCount()) == (expected_endpoint): : Protocol message serialized to a size different from what was originally expected. Perhaps it was modified by another thread during serialization?
terminate called after throwing an instance of 'google::protobuf::FatalException'
what(): CHECK failed: (output->ByteCount()) == (expected_endpoint): : Protocol message serialized to a size different from what was originally expected. Perhaps it was modified by another thread during serialization?
Aborted (core dumped)