HyRSM

HyRSM copied to clipboard

请教一下，训练时出现的问题

1

 我将ucf101的数据集规模缩小后，训练出现了上述问题，是数据集的问题吗？

li3256936

About data split

1

Hello authors,I download SSV2 from the official website，but the split of the data set looks different than yours. For example, the official data look like :"78687 54 19" ,that is...

Wickked-swag

About the data split

1

Hi, thanks for your excellent work. I have a question about the data split. You process the data in dataset into 8 frames per video, but I didn't find script...

Ambitious4

This is from [train_few_shot.txt](https://github.com/alibaba-mmai-research/HyRSM/blob/main/configs/projects/hyrsm/kinetics100/train_few_shot.txt) train0//train_256/air_drumming/-VtLx-mcPds_000012_000022.mp4 train0//train_256/air_drumming/-eGnPDG5Pxs_000053_000063.mp4 train0//train_256/air_drumming/-fTmHyOG-CY_000000_000010.mp4 train0//train_256/air_drumming/-ni1UWaBmL0_000056_000066.mp4 ... How to get these files?

smj700f

Set matching metric

2

Hello authors, I would like to ask where the code block about Set matching metric is located?

Liu-arch

Hi authors.When I run the code,it reports the following error.I can't solve it. Could you do me a favor? I tried to update the decord version to 0.6.0 but the...

Wickked-swag

How to read SSv2 videos?

1

Hello authors, it looks like you treat the SSv2 videos as .mp4 files ([here](https://github.com/alibaba-mmai-research/HyRSM/blob/main/datasets/base/ssv2_few_shot.py#L365)), but the SSv2 dataset is composed of .webm files. So have you done transformation before reading...

qizhongtan

Pre-trained checkpoints and features on MODEL_ZOO/FEATURES_ZOO missing

3

It looks like none of the links in the model/feature zoo pages link to anything. Do you have a plan to release them? UPDATE: are these them? https://github.com/alibaba/EssentialMC2/blob/main/MODEL_ZOO.md

beasteers

the NUM_TRAIN_TASKS and USE_CLASSIFICATION

4

Hi authors, I noticed that the NUM_TRAIN_TASKS under UCF setting is only 2500, is it enough to train the model? cuz I tried to reproduce your method, but the output...

yczhan

Some problem about training configs.

1

You offered different training configs for different datasets.  However, the same training config is used for all datasets in your code.   Which training config is right?

jiazheng-xing