Dataloader for UCF-CC-50

[jingliang95]

This is a very great work. Thanks a lot

Could you also share the dataloader and experimental setting (like crop size and so on) for dataset UCF-CC-50? Currently, I am working on this dataset, but I have trouble obtaining the results you reported. I think it may due to the dataloader or other experimental setting.

[jingliang95]

Specifically, I use UCF-CC-50 and do 5-fold cross validation. The crop size is 512 and I got average performance of 280.53 (MAE) and 419.0 (MSE), which is much worse than the reported number of 211.0 and 291.5. I think there must be something I missed. Could you help me with that? Thanks a lot

[Boyu-Wang]

The dataloader for UCF-CC-50 is the same as the dataloader for ShanghaiTech.
The crop size is 256.
Due to the limited number of images (50), the performance on UCF-CC-50 is less robust and may not reflect the true performance of the method. Even different ways of splitting affect the performance.
Those large-scale datasets (e.g. ShTech, QNRF, NWPU, JHU) are better benchmarks for evaluation.

[gabridego]

Hello @Boyu-Wang , is a model pretrained on JHU available ?

[lijinStudy]

The dataloader for UCF-CC-50 is the same as the dataloader for ShanghaiTech. The crop size is 256. Due to the limited number of images (50), the performance on UCF-CC-50 is less robust and may not reflect the true performance of the method. Even different ways of splitting affect the performance. Those large-scale datasets (e.g. ShTech, QNRF, NWPU, JHU) are better benchmarks for evaluation.

Hello, @Boyu-Wang, Could you share the dataloader for JHU dataset? Thanks!!!!