TypeError: 'ABCMeta' object is not subscriptable

[ssam2s]

Hello,

I’m trying to run the inference experiment for the UBFC-rPPG dataset using the unsupervised methods exactly as described in the README tutorial. In my UBFC-rPPG_UNSUPERVISED.yaml file, I updated DATA_PATH and CACHED_PATH to point to my local directories and set DO_PREPROCESS: True.

When I execute:
python main.py --config_file ./configs/infer_configs/UBFC_UNSUPERVISED.yaml

I immediately get the following error:
Traceback (most recent call last): File "main.py", line 11, in <module> from neural_methods import trainer File "/Data1/jhseo/rppg-toolbox/rPPG-Toolbox/neural_methods/trainer/__init__.py", line 2, in <module> import neural_methods.trainer.PhysnetTrainer File "/Data1/jhseo/rppg-toolbox/rPPG-Toolbox/neural_methods/trainer/PhysnetTrainer.py", line 9, in <module> from neural_methods.loss.PhysNetNegPearsonLoss import Neg_Pearson File "/Data1/jhseo/rppg-toolbox/rPPG-Toolbox/neural_methods/loss/PhysNetNegPearsonLoss.py", line 9, in <module> from torchvision import transforms File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torchvision/__init__.py", line 6, in <module> from torchvision import _meta_registrations, datasets, io, models, ops, transforms, utils File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torchvision/models/__init__.py", line 2, in <module> from .convnext import * File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torchvision/models/convnext.py", line 8, in <module> from ..ops.misc import Conv2dNormActivation, Permute File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torchvision/ops/__init__.py", line 1, in <module> from ._register_onnx_ops import _register_custom_op File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torchvision/ops/_register_onnx_ops.py", line 5, in <module> from torch.onnx import symbolic_opset11 as opset11 File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torch/onnx/__init__.py", line 46, in <module> from ._internal.exporter import ( # usort:skip. needs to be last to avoid circular import File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torch/onnx/_internal/exporter.py", line 42, in <module> from torch.onnx._internal.fx import ( File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torch/onnx/_internal/fx/__init__.py", line 1, in <module> from .patcher import ONNXTorchPatcher File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/torch/onnx/_internal/fx/patcher.py", line 11, in <module> import transformers # type: ignore[import] File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/transformers-4.51.3-py3.8.egg/transformers/__init__.py", line 26, in <module> from . import dependency_versions_check File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/transformers-4.51.3-py3.8.egg/transformers/dependency_versions_check.py", line 16, in <module> from .utils.versions import require_version, require_version_core File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/transformers-4.51.3-py3.8.egg/transformers/utils/__init__.py", line 24, in <module> from .backbone_utils import BackboneConfigMixin, BackboneMixin File "/Data1/jhseo/rppg-toolbox/lib/python3.8/site-packages/transformers-4.51.3-py3.8.egg/transformers/utils/backbone_utils.py", line 33, in <module> out_features: Optional[Iterable[str]], out_indices: Optional[Iterable[int]], stage_names: Optional[Iterable[str]] TypeError: 'ABCMeta' object is not subscriptable

Could you let me know how to resolve this TypeError: 'ABCMeta' object is not subscriptable?

I’ve attached my full config below for reference:
BASE: [''] TOOLBOX_MODE: "unsupervised_method" UNSUPERVISED: METHOD: ["ICA", "POS", "CHROM", "GREEN", "LGI", "PBV", "OMIT"] METRICS: ['MAE', 'RMSE', 'MAPE', 'Pearson', 'SNR', 'BA'] DATA: FS: 30 DATASET: UBFC-rPPG DO_PREPROCESS: True DATA_FORMAT: NDHWC DATA_PATH: "/Data1/jhseo/rppg-toolbox/rPPG-Toolbox/data/UBFC-rPPG" CACHED_PATH: "/Data1/jhseo/rppg-toolbox/rPPG-Toolbox/PreprocessedData" EXP_DATA_NAME: "" BEGIN: 0.0 END: 1.0 PREPROCESS: DATA_TYPE: ['Raw'] DATA_AUG: ['None'] LABEL_TYPE: Raw DO_CHUNK: False CHUNK_LENGTH: 180 CROP_FACE: DO_CROP_FACE: True BACKEND: 'HC' USE_LARGE_FACE_BOX: True LARGE_BOX_COEF: 1.5 DETECTION: DO_DYNAMIC_DETECTION: False DYNAMIC_DETECTION_FREQUENCY: 30 USE_MEDIAN_FACE_BOX: False RESIZE: H: 72 W: 72 INFERENCE: EVALUATION_METHOD: "FFT" EVALUATION_WINDOW: USE_SMALLER_WINDOW: False WINDOW_SIZE: 10

Thanks for your great work :)

[yahskapar]

Hi @ssam2s,

Did you make any modifications to the toolbox aside from the config you mentioned? The error message you got doesn't seem related to the config file you're trying to run for that matter, so I'm a bit confused and suspect the issue might either be with some other modification made to the toolbox code on your side or something going wrong with the environment setup.

If you try re-running the setup steps, do any errors appear in the terminal output? It's worth carefully inspecting this even if the setup appears to be successful based on the last few terminal printouts.

[ssam2s]

I discovered that the error above is caused by the version of the transformers package.
The type hints that used in versions later than 4.46.0 require Python 3.9 or higher, so they aren’t supported on Python 3.8.

To fix this, after completing the environment setup in the README, run:

pip uninstall transformers
pip install transformers==4.46.0

With this configuration, all of the TypeError issues disappear and the script runs normally.

[yahskapar]

I see, glad to hear this was resolved and thanks for sharing what solved it. I wonder if this is related to #373. I'll keep in mind what you observed when re-visiting that issue.

I'll go ahead and close this issue for now since it sounds like you solved your original problem and shared how.