Open yswang1717 opened 10 months ago
It would be really appreciated if you also share the required GPU memory to reproduce the paper performance.
Hello, Thank you for your interest in our work. We run all the experiments on a single RTX A5000 GPU with 24GB GPU memory. With the pre-extracted features and an extire A5000 GPU, it only takes several hours to run one single experiment.
Bests,
Thanks for the quick response!
Hello, have you successfully reached the author's experimental results? My accuracy is only 52%
Hello, have you successfully reached the author's experimental results? My accuracy is only 52%
Hello,
Do you run the experiments with our default setting?
Bests,
您好,请问您成功达到作者的实验结果了吗?我的准确率只有52%
你好,
您是否使用我们的默认设置运行实验?
最好的,
I ran train_incremental_ours.py. As shown above, here are the results of my experiment. The GPU I use is V100. Regarding audio and video feature extraction, I directly use the extracted features you shared.I refer to the settings in the paper. My experimental settings is shown below. I am very confused why the experimental result is only 52%. Hope to get your answer.Thank you so much.
您好,请问您成功达到作者的实验结果了吗?我的准确率只有52%
你好, 您是否使用我们的默认设置运行实验? 最好的,
I ran train_incremental_ours.py. As shown above, here are the results of my experiment. The GPU I use is V100. Regarding audio and video feature extraction, I directly use the extracted features you shared.I refer to the settings in the paper. My experimental settings is shown below. I am very confused why the experimental result is only 52%. Hope to get your answer.Thank you so much.
Thank you for your reply.
It seems that the hyperparameters of 'instance_contrastive', 'class_contrastive', and 'attn_score_distil' are all set to False in your experiment.
You can use our provided run_incremental_ours.sh to run your experiment directly. Please refer to sh run_incremental_ours.sh 'dataset'.
Bests,
Could you please share the required time for training in each experiment and relevant GPU information? Thanks !