Opting for SDD results actually produced ETH/UCY results

[SodaCoder]

Hello,

I think the data_process.py has some bugs, and even if I intend to train/test on SDD, it picks the saved .pkl files for ETH/UCY as the ADE/FDE results are similar to ETH/UCY.

Could you kindly look into this issue?

Thanks,
Sourav

[Gutianpei]

Have you saved the correct SDD pickles? You should only use the second half of the data_process.py for SDD, and the training script will automatically find the "SDD" directory to load training data

[SodaCoder]

Hello,

I copied the .pkl files from the Y-Net github repository and placed them inside the ./raw_data folder. I bypassed the code inside data_proeprocess.py for preprocessing the ETH/UCY dataset and used the later part of the code for preprocessing SDD only. Still, I am getting ADE/FDE scores close to ETH/UCY. Could you kindly look into this issue?

Thanks in advance,
Sourav