search

anosorae / IRRA

Cross-Modal Implicit Relation Reasoning and Aligning for Text-to-Image Person Retrieval (CVPR 2023)
MIT License
205 stars 27 forks source link

About ICFG-PEDES dataset #1

Closed ljwwwiop closed 1 year ago

ljwwwiop commented 1 year ago

Hi, I'm experiencing a memory shortage problem doing validation operations while training the ICFG-PEDES dataset, I'm using 12G of TITAN and have the training batch size set to 32 and the test batch size set to 128, but it seems that there is still a GPU memory shortage problem. I checked the training logs and found that there seems to be no val dataset, which seems to be the cause of the memory shortage. So I'm asking if the ICFG-PEDES data has a validation dataset loaded, or rather a json file of the cut data?

Thank you very much for open sourcing the code and I look forward to seeing the correct link to your paper, thanks!

ljwwwiop commented 1 year ago

image

anosorae commented 1 year ago

The SSAN author divided ICFG-PEDES into a training set and a testing set: the former comprises 34,674 image-text pairs of 3,102 persons, while the latter contains 19,848 image-text pairs for the remaining 1,000 persons. There is no val set for ICFG-PEDES datasets.

For the GPU memory shotage, I suggest you to further reduce the batchsize.

anosorae commented 1 year ago

And the correct link to our paper will be update after the camera-ready deadline of CVPR 2023, thank you for your attention to our work. @ljwwwiop

ljwwwiop commented 1 year ago

Thank you very much, I switched to 32G GPU and trained fine.