ipl-uw/2019-cvpr-aic-track-1-uwipl

Repository for 2019 CVPR AI City Challenge Track 1 from IPL@UW

2019-CVPR-AIC-Track-1-UWIPL

1. Single Camera Tracking

Please download the source code and follow the instruction from https://github.com/GaoangW/TNT/tree/master/AIC19.

2. Deep Feature Re_identification

Please download the source code and follow the instruction from https://github.com/ipl-uw/2019-CVPR-AIC-Track-2-UWIPL. Create video2img folder in the downloaded project (i.e., Video-Person-ReID/video2img/). Put and run python crop_img.py in the same folder in the downloaded dataset (i.e., aic19-track1-mtmc/test). You need to creat a folder track1_test_img in the same path (i.e., aic19-track1-mtmc/test/track1_test_img). After that, create a folder track1_sct_img_test_big and run python crop_img_big.py. Then, create a folder log in the dowanloaded project (i.e., Video-Person-ReID/log) and put the downloaded model file of track1 ReID in this folder. Finally, run python Graph_ModelDataGen.py to obtain the feature files (q_camids3_no_nms_big0510.npy, qf3_no_nms_big0510.npy and q_pids3_no_nms_big0510.npy).
The code is based on Jiyang Gao's Video-Person-ReID [code].

3. Trajectory-Based Camera Link Models

Put the feature files (q_camids3_no_nms_big0510.npy, qf3_no_nms_big0510.npy and q_pids3_no_nms_big0510.npy) in the Transition-Model folder of this project. Then, run python main_in_transition_matrix.py. Then, find the results in Transition-Model/transition_data /ICT-no_nms_big510/. The filename of the result should look like ict_greedy_iter1600_10.559_654_498.txt (1600 means the number of iteration, 10.559 represents the distance of embeddings at current iteration, 654 means the maximum globelID assigned to cross-camera vehicles, and 498 is the number of unique globelID of cross-camera vehicles after merging).

4. NMS

Put the output result (e.g. ict_greedy_iter1600_10.559_654_498.txt) from previous step in NMS folder. Then, run python NMS_filter.py to get the final result (e.g. ict_greedy_iter1600_10.559_654_498_big.txt) of Track1.

2019-cvpr-aic-track-1-uwipl's Issues

About "Trajectory Based Camera Link Models" on the slides

I can not understand these slide. Would you mind explaining them in more details? Or do you have a video of your presentation in the workshop?

Visualization of the MTMC tracking result

Would you mind telling me how to visualize the MTMC tracking result like the one you show on the CMT31_T1.pptx?

About dt_min and dt_max in camera_transitions.txt

Hello Sir,
I have a quick question about camera_transitions.txt in the second part of Track1. The last 2 columns were dt_min and dt_max. I guess it's the frame gap between zone_src and zone_dst. Am I right? If so, how did you get them?
I found your work very inspiring, any suggestion will be great.

Thanks

Model

Hello
Where can we found the model called track12_ta224_checkpoint_ep500.pth.tar that you mention in the README file that we should paste it into ./log/ folder ??
Thank you