Running inference on prithvi EO v2 models fails with terratorch predict ...
#438
Assignees
Labels
Comments
|
@paolofraccaro suggested to use this function to resize and pad the image to be ready for the model. |
|
With @romeokienzler suggestion. Config filedata:
class_path: terratorch.datamodules.GenericNonGeoSegmentationDataModule
init_args:
allow_substring_split_file: true
batch_size: 4
constant_scale: 1.0
dataset_bands:
- 2
- 1
- 0
- 3
- 4
- 5
drop_last: true
expand_temporal_dimension: false
ignore_split_file_extensions: true
img_grep: '*_S2GeodnHand.tif'
label_grep: '*_LabelHand.tif'
means:
- 0.12520133
- 0.13471393
- 0.107582
- 0.3236181
- 0.2341743
- 0.15878009
no_data_replace: 0.0
no_label_replace: -1
num_classes: 2
num_workers: 2
output_bands:
- 2
- 1
- 0
- 3
- 4
- 5
pin_memory: false
reduce_zero_label: false
rgb_indices:
- 2
- 1
- 0
stds:
- 0.07323416
- 0.06783548
- 0.07145836
- 0.09489725
- 0.07938496
- 0.07089546
test_data_root: /data/geodata-dbce399c854511efb3260a580a830dad/training_data
test_label_data_root: /data/geodata-dbce399c854511efb3260a580a830dad/labels
test_split: /data/geodata-dbce399c854511efb3260a580a830dad/split_files/test_data.txt
test_transform:
- class_path: albumentations.Resize
init_args:
always_apply: false
height: 512
interpolation: 1
p: 1.0
width: 512
- class_path: albumentations.pytorch.ToTensorV2
init_args:
always_apply: true
p: 1.0
transpose_mask: false
train_data_root: /data/geodata-dbce399c854511efb3260a580a830dad/training_data
train_label_data_root: /data/geodata-dbce399c854511efb3260a580a830dad/labels
train_split: /data/geodata-dbce399c854511efb3260a580a830dad/split_files/train_data.txt
train_transform:
- class_path: albumentations.Resize
init_args:
always_apply: false
height: 512
interpolation: 1
p: 1.0
width: 512
- class_path: albumentations.RandomCrop
init_args:
always_apply: false
height: 224
p: 1.0
width: 224
- class_path: albumentations.HorizontalFlip
init_args:
always_apply: false
p: 0.5
- class_path: albumentations.VerticalFlip
init_args:
always_apply: false
p: 0.5
- class_path: albumentations.pytorch.ToTensorV2
init_args:
always_apply: true
p: 1.0
transpose_mask: false
val_data_root: /data/geodata-dbce399c854511efb3260a580a830dad/training_data
val_label_data_root: /data/geodata-dbce399c854511efb3260a580a830dad/labels
val_split: /data/geodata-dbce399c854511efb3260a580a830dad/split_files/val_data.txt
val_transform:
- class_path: albumentations.Resize
init_args:
always_apply: false
height: 512
interpolation: 1
p: 1.0
width: 512
- class_path: albumentations.pytorch.ToTensorV2
init_args:
always_apply: true
p: 1.0
transpose_mask: false
deploy_config_file: true
model:
class_path: terratorch.tasks.SemanticSegmentationTask
init_args:
freeze_backbone: false
freeze_decoder: false
freeze_head: false
ignore_index: -1
loss: ce
lr: 0.001
model_args:
backbone: prithvi_eo_v2_300
backbone_bands:
- BLUE
- GREEN
- RED
- NIR_NARROW
- SWIR_1
- SWIR_2
backbone_pretrained: false
decoder: UperNetDecoder
decoder_channels: 256
decoder_scale_modules: true
head_dropout: 0.1
necks:
- indices:
- 5
- 11
- 17
- 23
name: SelectIndices
- name: ReshapeTokensToImage
- name: LearnedInterpolateToPyramidal
num_classes: 2
rescale: true
model_factory: EncoderDecoderFactory
output_most_probable: true
plot_on_val: 10
out_dtype: int16
seed_everything: 0
trainer:
accelerator: auto
accumulate_grad_batches: 1
barebones: false
check_val_every_n_epoch: 1
detect_anomaly: false
devices: auto
enable_checkpointing: true
fast_dev_run: false
inference_mode: true
log_every_n_steps: 10
logger: false
max_epochs: 2
max_steps: -1
num_nodes: 1
overfit_batches: 0.0
precision: 16-mixed
reload_dataloaders_every_n_epochs: 0
strategy: auto
sync_batchnorm: false
use_distributed_sampler: true
I get this error Box folder with artifacts: https://ibm.box.com/s/4nrf2jsfskjp5sbjg0ltrm5mh62f0pks |
|
With @paolofraccaro recomendation I get this error: |
|
Is there an updated I have checked the floods and burnscars hugging face repos, but nothing is there. |
|
HI, @WanjiruCate I saw the issue happens in the function I understand that it isn't always true. |
|
@WanjiruCate It runs, but I got a black image as output. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Describe the issue
Running inference with a finetuned model using the cli fails using an image downloaded using geostudio.
To Reproduce (optional, but appreciated)
Steps to reproduce the behavior:
Screenshots or log output (optional)
If applicable, add screenshots or log output to help explain your problem.
Log Output
Expected behavior (optional)
Prediction to happen seamlessly
Deployment information (optional)
Describe what you've deployed and how:
git+https://git@github.com/IBM/terratorch.git@1fa07cd806pip install git+https://git@github.com/IBM/terratorch.git@1fa07cd806The text was updated successfully, but these errors were encountered: