Adding padding at the input when necessary #342

Joao-L-S-Almeida · 2024-12-23T18:05:12Z

Should we add negative padding ? Could we remove some lines/rows instead of adding multiple lines/rows of zeroes ?
Are we interested in choosing the kind of padding or 'constant' is good enough ?

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

terratorch/models/pixel_wise_model.py

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

tests/resources/configs/manufactured-finetune_prithvi_swin_L.yaml

tests/resources/configs/manufactured-finetune_prithvi_vit_100.yaml

terratorch/models/encoder_decoder_factory.py

terratorch/models/pixel_wise_model.py

terratorch/models/scalar_output_model.py

terratorch/models/clay_model_factory.py

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

terratorch/models/backbones/select_patch_embed_weights.py

romeokienzler · 2025-01-24T11:07:19Z

@Joao-L-S-Almeida on my side the tests are failing, can you please have a look?

==================================================================================== 30 failed, 530 passed, 6 skipped, 327 warnings in 1352.63s (0:22:32) ====================================================================================

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Joao-L-S-Almeida · 2025-01-24T12:59:05Z

Are the errors looking like this ?

FAILED tests/test_backbones.py::test_out_indices_non_divisible[prithvi_eo_v1_100] - assert [2, 4, 8, 10] == (2, 4, 8, 10)
FAILED tests/test_backbones.py::test_out_indices_non_divisible[prithvi_eo_v2_300] - assert [2, 4, 8, 10] == (2, 4, 8, 10)
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_resnet[False-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_resnet[True-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_dofa[False-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_dofa[True-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-FCNDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-FCNDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-IdentityDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-IdentityDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-smp_Unet-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-smp_Unet-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-FCNDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-FCNDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-IdentityDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-IdentityDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-smp_Unet-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-smp_Unet-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-FCNDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-FCNDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-UperNetDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-UperNetDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-IdentityDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-IdentityDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-FCNDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-FCNDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-UperNetDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-UperNetDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-IdentityDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-IdentityDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_finetune.py::test_finetune_multiple_backbones[fit-prithvi_vit_100] - SystemExit: 2
FAILED tests/test_finetune.py::test_finetune_multiple_backbones[test-prithvi_vit_100] - SystemExit: 2
FAILED tests/test_finetune.py::test_finetune_multiple_backbones[validate-prithvi_vit_100] - SystemExit: 2
============================================================ 33 failed, 527 passed, 6 skipped, 166 warnings in 1536.95s (0:25:36) ============================================================
(terratorch)

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

terratorch/models/pixel_wise_model.py

blumenstiel · 2025-01-24T13:19:58Z

terratorch/models/encoder_decoder_factory.py

@@ -1,8 +1,8 @@
 # Copyright contributors to the Terratorch project

-
+from typing import List


It's better to use the types like list instead of the typing package. It's an old python package

@Joao-L-S-Almeida

terratorch/models/encoder_decoder_factory.py

terratorch/models/scalar_output_model.py

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Joao-L-S-Almeida · 2025-01-24T16:22:28Z

@blumenstiel The argument output_size was removed.

blumenstiel · 2025-01-24T17:52:44Z

tests/resources/configs/manufactured-finetune_prithvi_pixelwise_nondivisible.yaml

+      backbone: prithvi_eo_v2_300
+      # backbone_pretrained_cfg_overlay:
+      #   file: tests/prithvi_vit_300.pt
+      backbone_drop_path_rate: 0.3


Note that this parameter is called backbone_drop_path for prithvi ViT (following the name of the parameter in the ViT block). backbone_drop_path_rate is only working for prithvi swin.

is prithvi_eo_v2_300 ViT or swin? @blumenstiel

@romeokienzler Yes, all public pretrained models are ViT.

romeokienzler · 2025-01-24T20:30:14Z

terratorch/models/clay_model_factory.py

@@ -108,6 +109,7 @@ def build_model(

        # Path for accessing the model source code.
        self.syspath_kwarg = "model_sys_path"
+        backbone_kwargs, kwargs = extract_prefix_keys(kwargs, "backbone_")


@Joao-L-S-Almeida can you please update the docstring to explain the padding?

romeokienzler · 2025-01-24T20:32:07Z

terratorch/models/encoder_decoder_factory.py

@@ -128,6 +130,17 @@ def build_model(
        backbone_kwargs, kwargs = extract_prefix_keys(kwargs, "backbone_")
        backbone = _get_backbone(backbone, **backbone_kwargs)

+        # If patch size is not provided in the config or by the model, it might lead to errors due to irregular images.


@Joao-L-S-Almeida can you please update the docstring to explain the padding?

romeokienzler · 2025-01-24T21:51:10Z

Are the errors looking like this ?

FAILED tests/test_backbones.py::test_out_indices_non_divisible[prithvi_eo_v1_100] - assert [2, 4, 8, 10] == (2, 4, 8, 10)
FAILED tests/test_backbones.py::test_out_indices_non_divisible[prithvi_eo_v2_300] - assert [2, 4, 8, 10] == (2, 4, 8, 10)
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_resnet[False-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_resnet[True-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_dofa[False-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_dofa[True-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-FCNDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-FCNDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-IdentityDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-IdentityDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-smp_Unet-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-smp_Unet-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-FCNDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-FCNDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-IdentityDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-IdentityDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-smp_Unet-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-smp_Unet-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-FCNDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-FCNDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-UperNetDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-UperNetDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-IdentityDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[False-IdentityDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-FCNDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-FCNDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-UperNetDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-UperNetDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-IdentityDecoder-regression-expected0-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_dofa[True-IdentityDecoder-segmentation-expected1-dofa_large_patch16_224] - AttributeError: 'DOFA' object has no attribute 'embed_dim'
FAILED tests/test_finetune.py::test_finetune_multiple_backbones[fit-prithvi_vit_100] - SystemExit: 2
FAILED tests/test_finetune.py::test_finetune_multiple_backbones[test-prithvi_vit_100] - SystemExit: 2
FAILED tests/test_finetune.py::test_finetune_multiple_backbones[validate-prithvi_vit_100] - SystemExit: 2
============================================================ 33 failed, 527 passed, 6 skipped, 166 warnings in 1536.95s (0:25:36) ============================================================
(terratorch)

@Joao-L-S-Almeida

/home/romeokienzler/gitco/terratorch/.venv/lib64/python3.12/site-packages/lightning/pytorch/cli.py:526: LightningCLI's args parameter is intended to run from within Python like if it were from the command line. To prevent mistakes it is not recommended to provide both args and command line arguments, got: sys.argv[1:]=['tests/'], args=['test', '-c', 'tests/resources/configs/manufactured-finetune_prithvi_pixelwise_pad.yaml'].

tests/test_finetune.py::test_finetune_pad[validate-prithvi_eo_v2_300]
/home/romeokienzler/gitco/terratorch/.venv/lib64/python3.12/site-packages/lightning/pytorch/cli.py:526: LightningCLI's args parameter is intended to run from within Python like if it were from the command line. To prevent mistakes it is not recommended to provide both args and command line arguments, got: sys.argv[1:]=['tests/'], args=['validate', '-c', 'tests/resources/configs/manufactured-finetune_prithvi_pixelwise_pad.yaml'].

tests/test_finetune.py::test_finetune_metrics_from_file[prithvi_swin_B]
/home/romeokienzler/gitco/terratorch/.venv/lib64/python3.12/site-packages/lightning/pytorch/cli.py:526: LightningCLI's args parameter is intended to run from within Python like if it were from the command line. To prevent mistakes it is not recommended to provide both args and command line arguments, got: sys.argv[1:]=['tests/'], args=['fit', '-c', 'tests/resources/configs/manufactured-finetune_prithvi_swin_B_metrics_from_file.yaml'].

tests/test_generic_dataset.py::TestGenericSegmentationDataset::test_data_type_regression_float_long
tests/test_generic_dataset.py::TestGenericSegmentationDataset::test_data_type_regression_float_long
/home/romeokienzler/gitco/terratorch/.venv/lib64/python3.12/site-packages/rioxarray/_io.py:1136: NotGeoreferencedWarning: Dataset has no geotransform, gcps, or rpcs. The identity matrix will be returned.
warnings.warn(str(rio_warning.message), type(rio_warning.message)) # type: ignore

tests/test_prithvi_model_factory.py::test_create_pixelwise_model[UperNetDecoder-regression-expected0-prithvi_eo_v1_100]
tests/test_prithvi_model_factory.py::test_create_pixelwise_model[UperNetDecoder-segmentation-expected1-prithvi_eo_v1_100]
tests/test_prithvi_model_factory.py::test_create_pixelwise_model_no_in_channels[UperNetDecoder-regression-expected0-prithvi_eo_v1_100]
tests/test_prithvi_model_factory.py::test_create_pixelwise_model_no_in_channels[UperNetDecoder-segmentation-expected1-prithvi_eo_v1_100]
tests/test_prithvi_model_factory.py::test_create_pixelwise_model_with_aux_heads[UperNetDecoder-regression-expected0-prithvi_eo_v1_100]
tests/test_prithvi_model_factory.py::test_create_pixelwise_model_with_aux_heads[UperNetDecoder-segmentation-expected1-prithvi_eo_v1_100]
tests/test_prithvi_model_factory.py::test_create_pixelwise_model_with_extra_bands[UperNetDecoder-regression-expected0-prithvi_eo_v1_100]
tests/test_prithvi_model_factory.py::test_create_pixelwise_model_with_extra_bands[UperNetDecoder-segmentation-expected1-prithvi_eo_v1_100]
/home/romeokienzler/gitco/terratorch/terratorch/models/decoders/upernet_decoder.py:37: UserWarning: DeprecationWarning: scale_modules is deprecated and will be removed in future versions. Use LearnedInterpolateToPyramidal neck instead.
warnings.warn(

-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
========================================================================================================== short test summary info ===========================================================================================================
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_resnet[False-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_classification_model_resnet[True-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-FCNDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-FCNDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-IdentityDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-IdentityDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-smp_Unet-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[False-smp_Unet-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-FCNDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-FCNDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-IdentityDecoder-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-IdentityDecoder-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-smp_Unet-regression-expected0-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
FAILED tests/test_encoder_decoder_torchgeo_models.py::test_create_pixelwise_model_resnet[True-smp_Unet-segmentation-expected1-ssl4eos12_resnet50_sentinel2_all_decur] - AttributeError: 'ResNet' object has no attribute 'forward_intermediates'
==================================================================================== 14 failed, 532 passed, 6 skipped, 327 warnings in 757.23s (0:12:37) =====================================================================================

Joao-L-S-Almeida · 2025-01-24T22:24:24Z

@romeokienzler It seems to be torchgeo version trickery. It just works with the version from the requirements: https://github.com/microsoft/torchgeo.git@fedf99375535f801565856cd774bfa9e5a251d55.

Signed-off-by: Joao Lucas de Sousa Almeida <[email protected]>

romeokienzler · 2025-01-27T08:48:19Z

@Joao-L-S-Almeida sorry for the late reply, my build/test queue is fill with other branches, but will test before your morning...

Adding padding at the input when necessary

842941b

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Joao-L-S-Almeida self-assigned this Dec 23, 2024

Joao-L-S-Almeida linked an issue Dec 23, 2024 that may be closed by this pull request

Add padding while evaluating on the original mask size #261

Open

Joao-L-S-Almeida mentioned this pull request Dec 23, 2024

Add padding while evaluating on the original mask size #261

Open

blumenstiel reviewed Dec 27, 2024

View reviewed changes

terratorch/models/pixel_wise_model.py Outdated Show resolved Hide resolved

blumenstiel reviewed Dec 27, 2024

View reviewed changes

terratorch/models/pixel_wise_model.py Outdated Show resolved Hide resolved

Joao-L-S-Almeida added 8 commits January 2, 2025 14:21

patch_size as a explicit argument for PixelWiseModel

6528861

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

logging

8376a5e

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Cropping image

8fa3bba

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

cropping image for scaler model

6fb8c95

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

patch_size could be None

5f37ba7

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Adapting the Clay factory to support patch_size and minor adjusts

5cb27dc

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Trying to reduce the cost of these tests

ba43134

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

pad_images must be in utils.py

9c26eab

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Joao-L-S-Almeida marked this pull request as ready for review January 3, 2025 18:26

Joao-L-S-Almeida changed the title ~~[WIP] Adding padding at the input when necessary~~ Adding padding at the input when necessary Jan 3, 2025

Joao-L-S-Almeida requested review from blumenstiel and romeokienzler January 3, 2025 18:26

blumenstiel reviewed Jan 6, 2025

View reviewed changes

tests/resources/configs/manufactured-finetune_prithvi_swin_L.yaml Show resolved Hide resolved

blumenstiel reviewed Jan 6, 2025

View reviewed changes

tests/resources/configs/manufactured-finetune_prithvi_vit_100.yaml Outdated Show resolved Hide resolved

blumenstiel reviewed Jan 6, 2025

View reviewed changes

terratorch/models/encoder_decoder_factory.py Outdated Show resolved Hide resolved

blumenstiel reviewed Jan 6, 2025

View reviewed changes

terratorch/models/pixel_wise_model.py Show resolved Hide resolved

blumenstiel reviewed Jan 6, 2025

View reviewed changes

terratorch/models/scalar_output_model.py Show resolved Hide resolved

blumenstiel reviewed Jan 6, 2025

View reviewed changes

terratorch/models/clay_model_factory.py Outdated Show resolved Hide resolved

Joao-L-S-Almeida added 6 commits January 6, 2025 10:03

Cropping images could be a necessary operation

c4fd736

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

The cropping must be placed before the head in case of scalar models

b70a368

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Creating extra images for tests

ecca3aa

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Minor changes

e09cd79

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

img_size also could be necessary

6fdf1b7

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

conditional cropping

6178b47

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Joao-L-S-Almeida added 3 commits January 21, 2025 10:26

Cropping the image when necessary

6786612

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Merge branch 'add/pad' of github.com:IBM/terratorch into add/pad

f4fa6a0

tests no more required

bc68b87

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

fmartiescofet mentioned this pull request Jan 21, 2025

Feat: Implement multiple test dataloaders in all tasks #330

Merged

romeokienzler reviewed Jan 24, 2025

View reviewed changes

terratorch/models/backbones/select_patch_embed_weights.py Outdated Show resolved Hide resolved

romeokienzler reviewed Jan 24, 2025

View reviewed changes

terratorch/models/backbones/select_patch_embed_weights.py Show resolved Hide resolved

merging with main

5265a30

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Updating model name

17a608e

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

blumenstiel reviewed Jan 24, 2025

View reviewed changes

terratorch/models/pixel_wise_model.py Outdated Show resolved Hide resolved

blumenstiel reviewed Jan 24, 2025

View reviewed changes

terratorch/models/encoder_decoder_factory.py Outdated Show resolved Hide resolved

blumenstiel reviewed Jan 24, 2025

View reviewed changes

terratorch/models/scalar_output_model.py Outdated Show resolved Hide resolved

Joao-L-S-Almeida added 2 commits January 24, 2025 10:51

Fixing indent

1ff8938

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Removing output_size

b0a4780

Signed-off-by: João Lucas de Sousa Almeida <[email protected]>

Joao-L-S-Almeida requested review from blumenstiel and romeokienzler January 24, 2025 16:41

blumenstiel reviewed Jan 24, 2025

View reviewed changes

romeokienzler reviewed Jan 24, 2025

View reviewed changes

indent

0dc1e95

Signed-off-by: Joao Lucas de Sousa Almeida <[email protected]>

Joao-L-S-Almeida requested review from blumenstiel and romeokienzler January 28, 2025 12:29

Joao-L-S-Almeida added 0.99.8 and removed 1.0 labels Jan 28, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Adding padding at the input when necessary #342

Adding padding at the input when necessary #342

Joao-L-S-Almeida commented Dec 23, 2024

romeokienzler commented Jan 24, 2025

Joao-L-S-Almeida commented Jan 24, 2025

blumenstiel Jan 24, 2025

romeokienzler Jan 24, 2025

Joao-L-S-Almeida commented Jan 24, 2025

blumenstiel Jan 24, 2025

romeokienzler Jan 24, 2025

blumenstiel Jan 27, 2025

romeokienzler Jan 24, 2025

romeokienzler Jan 24, 2025

romeokienzler commented Jan 24, 2025

Joao-L-S-Almeida commented Jan 24, 2025

romeokienzler commented Jan 27, 2025

		@@ -1,8 +1,8 @@
		# Copyright contributors to the Terratorch project


		from typing import List

Adding padding at the input when necessary #342

Are you sure you want to change the base?

Adding padding at the input when necessary #342

Conversation

Joao-L-S-Almeida commented Dec 23, 2024

romeokienzler commented Jan 24, 2025

Joao-L-S-Almeida commented Jan 24, 2025

blumenstiel Jan 24, 2025

Choose a reason for hiding this comment

romeokienzler Jan 24, 2025

Choose a reason for hiding this comment

Joao-L-S-Almeida commented Jan 24, 2025

blumenstiel Jan 24, 2025

Choose a reason for hiding this comment

romeokienzler Jan 24, 2025

Choose a reason for hiding this comment

blumenstiel Jan 27, 2025

Choose a reason for hiding this comment

romeokienzler Jan 24, 2025

Choose a reason for hiding this comment

romeokienzler Jan 24, 2025

Choose a reason for hiding this comment

romeokienzler commented Jan 24, 2025

Joao-L-S-Almeida commented Jan 24, 2025

romeokienzler commented Jan 27, 2025