uark-aicv / tapg-agentenvinteration Goto Github PK

I have tried to implement your code on my custom dataset.May i ask what model you use for extract 2048 dim video C3D feature?i have look into C3D and 3D resnet pretrained model and it doesn't have layer of 2048 dim output(which i find is 512,4096 etc),if i change the model layer it means i cant use pretrained model,have you use some post processing on the feature output or can you give me some guide of the feature extraction how i can get my own video feature for the training

hello,iam very confused about the evaluate during training.i have prepare the feature data and my annotation file of the dataset,and start training, however i have no idea how can i generate the proposal_file and ground_truth_file for the evaluate part in the training loop,as i just start training i dont have a result.json,and because of that i got error being lack of files for evaluate during training,can you explain the data prepration for the training?How can i solve this problem?
https://github.com/UARK-AICV/TAPG-AgentEnvInteration/blob/c62acedc7f988b8f78e7ba8ee15f1e1f16cef5a1/main.py#L140C1-L140C67