search

realcrane / Human-Trajectory-Prediction-via-Neural-Social-Physics

Our ECCV 2022 paper Human Trajectory Prediction via Neural Social Physics
127 stars 19 forks source link

Question about the total number of SDD train and test datasets #13

Open jiajiaxiaoskx opened 1 year ago

jiajiaxiaoskx commented 1 year ago

Hi. I would like to thank you first for sharing your great work!

I have two questions about the SDD train and test data. 1、the previous work ( including Ynet you mentioned in the paper) use 17 scenes data as test data，but in your test_masks folder，it seems that you use 16 scenes data ( I also check the scene number in test.pickle ). I find that you leave the nexus_5 scene data, which is different from Ynet. 2、I run the two script in SDD_ini folder to generate the SDD train data and test for the network input. However, when I check the total number of trajectories, I find that after such preprocess, the number of trajectories in test data is 2123( the number after prepocessing in Ynet is 2829). It seems that you delete some trajectories during process in [generate_train_test_data_SDD.py], which may influence (improve) the testing ade/fde. Could you please answer why you do that? Thanks a lot!

JiangbeiYue commented 1 year ago

Hi, thanks for your question.

1 The processing of raw data in our method and Y-net (previous works) is different, so their final processed data tend to be different. We didn’t find usable data in the nexus_5 scene after processing data in our method. Our way of processing data is based on that of Social GAN, while Y-net used the pre-processed SDD data from TrajNet benchmark. The differences between the two ways of processing data in our method and Y-net include the window size, the treatment for outliers and etc. You can find all details of our data processing in our data processing code (data/SDD_ini/generate_train_test_data_SDD.py). Details of data processing of Y-net may be found in their website.

2 We did not delete any trajectories. Could you specify where (line numbers) in generate_train_test_data_SDD.py you found that trajectories were deleted? As mentioned before, the processing of raw data in our method and Y-net is different, so their final processed data tend to be different. Our way of processing data is based on that of Social GAN, while Y-net used the pre-processed SDD data from TrajNet benchmark.