Results Reproduce

[jakeiaCyber]

Hi, @bjzhb666

Thank you for releasing this code.

I am currently facing some problems while reproducing the "Continual Forgetting Results for Face Recognition", and I hope you can help clarify them.

I am using NVIDIA GeForce RTX 3090, and cuda version is 11.4.

I follow the instructions in the README and execute run_cl_forget.sh. I review the run_cl_forget.sh and observe that the following code was executed as shown in the image.

The following image shows the Task 0 section of my log. The results show a significant difference compared to TABLE 3 of the "Practical Continual Forgetting for Pre-trained Vision Models" paper.

I hope you can provide me with some guidance on reproducing the GS-LoRA results in the "Continual Forgetting Results for Face Recognition."

I would greatly appreciate it. Once again, thank you for releasing this excellent work.

Here is my log.

output.log

[bjzhb666]

Hi, @bjzhb666. Thank you so much for releasing the code. I was using your pre trained model checkpoint "Backbone_VIT_Epoch_1185_Batch_45020_Time_2024-09-26-03-26_checkpoint.pth " for face recognition task using your specified dataset Casia 100.

After using this i tested the pre-trained model for the Face Transformer. To my surprise i am getting very low accuracy of around 48.553%. Can you provide me some guidance regarding the same. Are you ignoring some classes or is this the same accuracy that you are getting.

I guess you did something wrong, the checkpoint is correct.

[bjzhb666]

Hi, @bjzhb666

Thank you for releasing this code.

I am currently facing some problems while reproducing the "Continual Forgetting Results for Face Recognition", and I hope you can help clarify them.

I am using NVIDIA GeForce RTX 3090, and cuda version is 11.4.

I follow the instructions in the README and execute run_cl_forget.sh. I review the run_cl_forget.sh and observe that the following code was executed as shown in the image.

The following image shows the Task 0 section of my log. The results show a significant difference compared to TABLE 3 of the "Practical Continual Forgetting for Pre-trained Vision Models" paper. I hope you can provide me with some guidance on reproducing the GS-LoRA results in the "Continual Forgetting Results for Face Recognition."

I would greatly appreciate it. Once again, thank you for releasing this excellent work.

Here is my log.

output.log

I will check the hyperparameters, thanks for pointing it out.