search

OpenGVLab / UniFormerV2

[ICCV2023] UniFormerV2: Spatiotemporal Learning by Arming Image ViTs with Video UniFormer
https://arxiv.org/abs/2211.09552
Apache License 2.0
291 stars 18 forks source link

dataloader长度为0 #70

Open hahapt opened 6 months ago

hahapt commented 6 months ago

我采用./exp/k400/k400_b16_f8x224/run.sh作为基础配置，数据集选用自有数据集。在config.yaml中已经将num_classes更改为2，且train/val/test.csv中按照0/1来区分视频类别。在kinetics_sparse.py的_construct_loader函数中已经验证过可以正确读取label和video_path，且video_path已校验过路径正确存在目标，但dataloader最终读取长度为0。

Andy1621 commented 6 months ago

可以输出一些_construct_loader的中间变量看看