chinmaynehate / dfspot-deepfake-recognition Goto Github PK

Thanks a lot for this wonderful code-base. Is it possible to get the training script? I want to check the performance of the proposed method on custom dataset after training?

Hello,

I want to express my gratitude for your efforts. I've encountered an issue when attempting to run both ViT and ViTST models. Specifically, I'm encountering an error when executing the following command:

python spot_deepfakes.py --media_type video --data_dir /DFSpot-Deepfake-Recognition/sample_videos/dfdc/fake/ --dataset dfdc --model ViT --model_dir /DFSpot-Deepfake-Recognition/models --annotate True --output_dir /DFSpot-Deepfake-Recognition/src/output --video_id 0

The error message I'm receiving is as follows:

RuntimeError: mat1 and mat2 shapes cannot be multiplied (300x201728 and 1024x1)

Interestingly, I am able to run the same command successfully with both TimmV2 and Timmvst models. However, I face this error exclusively when attempting to run ViT and ViTST models. It's worth noting that the video input remains consistent across all four models.
Any assistance with resolving this issue would be greatly appreciated.

When I am trying to run this command:
python3 spot_deepfakes.py --media_type video --data_dir ../sample_videos/dfdc/fake/ --dataset dfdc --model TimmV2 TimmV2ST ViT ViTST --model_dir ../models/ --video_id 0 1 2 --annotate True --device 0 --output_dir output/

I am receiving the following error:
Traceback (most recent call last):
File "/home/sanskriti/df_spot/DeepFake-Spot/src/spot_deepfakes.py", line 121, in
main()
File "/home/sanskriti/df_spot/DeepFake-Spot/src/spot_deepfakes.py", line 99, in main
predictions = utils.extract_predict_annotate(output_dir, ensemble_models, video_glob, video_idxs, transformer, blazeface_dir, device, models_loaded)

File "/home/sanskriti/df_spot/DeepFake-Spot/src/utils/utils.py", line 343, in extract_predict_annotate
vid = cv2.VideoCapture(video_glob[ne])
~~~~~~~~~~^^^^
IndexError: list index out of range

It is similar to the other datasets as well.