The directory structure of the WorldExpo'10 dataset

[Xluo-lc]

The directory structure of the WorldExpo'10 dataset

[maheshkkumar]

Hi,

Here is an overview of the directory structure for the test split containing 5 scenes for WorldExpo'10 dataset.

├─WorldExpo
├──test
├───csvs
├─────104207
├─────200608
├─────200702
├─────202201
├─────500717
├───frames
├─────104207
├─────200608
├─────200702
├─────202201
├─────500717

• csvs: This directory contains scene-specific density maps for every input image as a .csv file.
• frames: This directory contains the input RGB images grouped into their respective scenes.

Follow the same setup for the train split.

I hope this helps.

[Xluo-lc]

Thank you. May I ask how you solve the problem of insufficient memory?I can't use multiple GPUs (nn.DataParallel) directly in you code.
RuntimeError: CUDA out of memory. Tried to allocate 40.00 MiB (GPU 0; 12.00 GiB total capacity; 10.43 GiB already allocated; 25.12 MiB free; 134.05 MiB cached)

[maheshkkumar]

This version of the code supports training only on a single GPU. However, to make it compatible with multi-GPU training, some functionalities in the dataloader needs to be changed. Since there is no requirement for this now, we do not have plans to update the codebase. Feel free to reopen this, if there are other concerns. Thanks!

[chenwenhe11]

Hi, how did you get the WorldExpo'10 dataset? I sent an email to Prof. Xie Rong but never got a reply.
If possible, could you share the download link for the dataset with me, thank you very much!
My email is [email protected].