Closed menghuaa closed 2 years ago
menghuaa
commented
2 years ago 除了这个问题，我还想问下，你们在将activitynet1.3的特征进行线性插值时，有没有考虑它的帧频变成了25fps呢？因为线性插值的代码中会有计算num_frame和feature_frame的地方。num_frame（视频总帧数）你们是直接取标注json文件中的duration_frame值，还是使用视频的持续时间*25得到呢？因为标注文件中的帧频并不都是25fps，而我们使用的原始特征是以25fps的帧频提取特征。我看您anet.py中的代码好像并没有考虑25fps导致视频总帧数变化这个问题，而是直接采用标注json文件中的duration_frame值。
JackyTown
commented
2 years ago I can run the code on my machine.
You can re-scale the feature to 100 with https://github.com/JJBOY/BMN-Boundary-Matching-Network/blob/master/data/activitynet_feature_cuhk/data_process.py.
I do not know the feature FPS, since we directly use the feature of https://github.com/PeisenZhao/Bottom-Up-TAL-with-MR/issues/5.
Although we do not know the feature FPS, feature_frames == num_frames // 16 * 16 We perform interpolation based on the num_frames of json.
menghuaa
commented
2 years ago I can run the code on my machine.
You can re-scale the feature to 100 with https://github.com/JJBOY/BMN-Boundary-Matching-Network/blob/master/data/activitynet_feature_cuhk/data_process.py.
I do not know the feature FPS, since we directly use the feature of PeisenZhao/Bottom-Up-TAL-with-MR#5.
Although we do not know the feature FPS, feature_frames == num_frames // 16 * 16 We perform interpolation based on the num_frames of json.
can you provide the tsn feature of thumos14?
您好，我去年复现了rtd 在thumos14数据集上的代码，但是我用同样的环境运行anet的代码，却发生了环境方面的错误，nccl278 error。刚才我发现anet代码中util/misc.py文件中，这个函数torch.distributed.init_process_group( backend=args.dist_backend, init_method=args.dist_url, world_size=args.world_size, rank=args.rank, )，最后一个参数后面有一个逗号（rank=args.rank,），但是thumos14的代码中是没有，我想问下这个多出来的逗号是一个错误吗？