search

AshwinPrksh00 / Adapt-FuseNet

Repository for the implementation of the paper "Adapt-FuseNet: Context-aware Multimodal Fusion of Face and Gait Features using Attention Techniques for Human Identification"
Apache License 2.0
1 stars 1 forks source link

Training process not clear #4

Open kailashhambarde opened 5 months ago

kailashhambarde commented 5 months ago

I think the concept behind this paper involves training a single model that can extract both facial and gait features, and then you used AdaFuseNet and so on...... If I understand correctly, did you train the facial and gait feature extraction separately?

I noticed that you have two files, "train_face.py" and "train_gait.py," which confused me. I assumed that you had a single training loop.

AshwinPrksh00 commented 5 months ago

Hi @kailashhambarde, You have assumed correctly. Face and Gait are trained separately to get the model weights as a single loop training required more computation power that what we were using. These model weights are then plugged into the AdaptFuseNet model, and using them, the fusion module would be trained.

kailashhambarde commented 5 months ago

Could you please help me find the training code for AdaFuseNet in your repository?

AshwinPrksh00 commented 5 months ago

The training process for Adapt-FuseNet would be uploaded after journal publication of the paper. I would suggest to wait till the publication.

kailashhambarde commented 5 months ago

Thank you for your response. I have no issues and wish you the best of luck with your publication.

I just have one more question: Can you provide me with information on how the testing and deployment for the model can be done?

Specifically, I would like to know if my understanding is correct that the image/frame is first passed through the face and gait models before being passed to the adafusenet model.

Also, I would like to know about the time complexity of your overall model, considering that images per frame have to pass through three models (face, gait, and adafusenet).

AshwinPrksh00 commented 4 months ago

@kailashhambarde Sorry for taking so long before replying. As for your questions, I'll answer in bullet points for each of your questions so that it's easier to understand and collate.