Hi, and thank you for releasing APGCC - great work at ECCV 2024.
We are a group of MSc students at the Technical University of Denmark (DTU) working on a course project on multi-resolution crowd counting. We are building a benchmarking pipeline to evaluate how existing crowd counting models degrade under synthetic resolution reduction (2x, 4x downscaling) on the NWPU-Crowd dataset.
We would like to include APGCC in our comparison alongside DM-Count and CLIP-EBC. However, we noticed that only the ShanghaiTech Part A checkpoint has been released publicly. Training APGCC ourselves on NWPU would introduce a data contamination issue for our benchmarking, since we evaluate on the NWPU validation set.
Would it be possible to share the pretrained NWPU-Crowd checkpoint? Even an informal share via Google Drive would be very helpful. We are happy to cite your work and acknowledge the checkpoint in our report.
Thank you very much for considering this!
Hi, and thank you for releasing APGCC - great work at ECCV 2024.
We are a group of MSc students at the Technical University of Denmark (DTU) working on a course project on multi-resolution crowd counting. We are building a benchmarking pipeline to evaluate how existing crowd counting models degrade under synthetic resolution reduction (2x, 4x downscaling) on the NWPU-Crowd dataset.
We would like to include APGCC in our comparison alongside DM-Count and CLIP-EBC. However, we noticed that only the ShanghaiTech Part A checkpoint has been released publicly. Training APGCC ourselves on NWPU would introduce a data contamination issue for our benchmarking, since we evaluate on the NWPU validation set.
Would it be possible to share the pretrained NWPU-Crowd checkpoint? Even an informal share via Google Drive would be very helpful. We are happy to cite your work and acknowledge the checkpoint in our report.
Thank you very much for considering this!