search

USTC-Video-Understanding / I3D_Finetune

TensorFlow code for finetuning I3D model on UCF101.
144 stars 43 forks source link

Trained Models #18

Open ovesreinier opened 5 years ago

ovesreinier commented 5 years ago

Hello Rhythmblue

I am traying to use your pretrained models in UCF101. In this site I found 2 links to the models (google drive and the other in Baidu) The google drive version is in you trash, can you restore one more time please ?

Regards, Reinier

vra commented 5 years ago

Hi Reinier, The models have been recovered. Please check again.

ovesreinier commented 5 years ago

Dear Yunfeng I would like to thank you for sharing me the training models. But I am having a trouble, when I test the first split of UCF101 dataset with your test.py code "CUDA_VISIBLE_DEVICES=0 python test.py ucf101 mixed 1" I only reach 0.77 accuracy and in your paper the accuracy is higher than this (> 0.97)

Can you tell me what can I do for achieve your accuracy (I am using train/test splits from the original database)

Best regards Reinier and sorry for delay

On Thu, Dec 13, 2018 at 11:24 PM Yunfeng Wang notifications@github.com wrote:

Hi Reinier, The models have been recovered. Please check again.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/USTC-Video-Understanding/I3D_Finetune/issues/18#issuecomment-447217557, or mute the thread https://github.com/notifications/unsubscribe-auth/AVDo_mZQtX3YTYS7FXHtr_yVsoCiSDIIks5u4zYrgaJpZM4ZSAub .

anhnguyen0713 commented 5 years ago

Dear Yunfeng, I also encountered the same issue when testing the model with mixed modality. I cannot reach 0.97 accuracy as you did. Could you tell me how you achieved that result? I really appreciate that.

Kind regards,

sch3 commented 5 years ago

I am also encountering the same issue with mixed modality. Please advise on how the reported result was achieved.