search

nwpu-zxr / VadCLIP

VadCLIP official Pytorch implementation
Apache License 2.0
115 stars 11 forks source link

Question About Test Results Obtained Using Provided Pretrained Weights #3

Closed mycroft1603 closed 10 months ago

mycroft1603 commented 10 months ago

Thanks for your time and for sharing your awesome work! I've conducted tests using the provided pre-trained models and pre-extracted features, and it seems that the AUC in coarse-grained scenarios is lower than the results reported in the paper. If there some crucial steps I overlooked, or there are additional calculations required to achieve the results mentioned in the paper?

test
nwpu-zxr commented 10 months ago

We have verified that the test results of the given features and model parameters match the results in the paper under two different devices. Please check if the ucf_option.py has been changed. Also, please ensure that the Python version is 3.8 or higher and the PyTorch version is 2.0 or higher.

GP0123 commented 7 months ago

I use the Python version is 3.8 and PyTorch version is 2.0.0, but the auc is that 1713676052602 it doesn't up to the result of the paper