Experimental results about adapt

[haoshuai714]

How many video frames are you using? It feels like you used 4 or 8 frames, and the paper used 32 frames.
I'm not the author, I'm just guessing.

[LPeng625]

How many video frames are you using? It feels like you used 4 or 8 frames, and the paper used 32 frames. I'm not the author, I'm just guessing.
Thank you. Have you reproduced the results in the paper?

[jxbbb]

We use an early stop strategy, and you can check if the checkpoint before 40 is better.

[jxbbb]

btw, can I take a look at the Tensorboard log?

[LPeng625]

btw, can I take a look at the Tensorboard log?

Thanks for the reply. There are indeed better results before 40, but there is still a big gap with paper.

[LPeng625]

ow many video frames are you using? It feels like you used 4 or 8 frames, and the paper used 32 frames.
I'm not the author, I'm just guessing.

32, this result seems to use 4 or 8 frames. We set bs to 2, I don't know if this is the reason.

[LPeng625]

We use an early stop strategy, and you can check if the checkpoint before 40 is better.

Thanks, the best results are really close to the paper. I guess the gap is related to the training environment and the setting of the bs.

[ljy2222]

We use an early stop strategy, and you can check if the checkpoint before 40 is better.

Thanks, the best results are really close to the paper. I guess the gap is related to the training environment and the setting of the bs.

Hi, I met the same problem, have you successfully reproduced the results in the paper, i.e. 32 frames in Table VI.

[LPeng625]

We use an early stop strategy, and you can check if the checkpoint before 40 is better.

Thanks, the best results are really close to the paper. I guess the gap is related to the training environment and the setting of the bs.

Hi, I met the same problem, have you successfully reproduced the results in the paper, i.e. 32 frames in Table VI.

Hello, there is a gap between the results of our experiment and the paper results.

[ananditajam]

Hi @jxbbb, the link(https://pan.baidu.com/s/1_eW-kLVBhf8lWGObGa4i9A?pwd=7zrz) you provided for downloading the preprocessed dataset is inaccessible. Could you please send the dataset via email?