search

chaoyuaw / pytorch-coviar

Compressed Video Action Recognition
https://www.cs.utexas.edu/~cywu/projects/coviar/
GNU Lesser General Public License v2.1
500 stars 126 forks source link

only get 57% acc on mv vector, ucf101 split1 #89

Open DawnChou opened 3 years ago

DawnChou commented 3 years ago

Hey, I just copied the options and settings when I trained coviar on UCF101-split1, but I only got 57% acc (validation) when training. Is there something wrong? Any help? Thanks.

villawang commented 3 years ago

Are you using the training script? Try testing the script. In the testing, they used more segments and also 10 crops as augmentation.

JunLiangZ commented 2 years ago

Hello, can you send me the "coviar" package, I think there is a problem with my installation, thank you, my email is 1789163036@qq.com

Wendy-up commented 8 months ago

Are you using the training script? Try testing the script. In the testing, they used more segments and also 10 crops as augmentation.

Hey, I also followed the options and settings in this project on HMDB51, but got very low accuracy compared with the results in the paper, and I noticed that in the paper the author mentioned that their models were pretrained on ImageNet. so dose this mean that I have to pretrain the model on ImageNet to get similar results with the paper, or just because there is something else wrong so that I get the poor results? Really thanks!