swetanjal / mussp-efficient-min-cost-flow-algorithm-for-multi-object-tracking Goto Github PK

View Code? Open in Web Editor NEW

An efficient algorithm for solving data association problems modelled as a Minimum cost flow problem applied in the field of Multi Object Tracking

C++ 11.40% Python 87.26% Shell 1.34%

mussp-efficient-min-cost-flow-algorithm-for-multi-object-tracking's Introduction

muSSP-Efficient-Min-cost-Flow-Algorithm-for-Multi-object-Tracking

Team Members:

Teja Dhondu
Swetanjal Dutta
Nishanth Sharma

Run demo as follows:

>> python3 create_graph.py input_detections_folder graph_file
>> ./a.out -i graph_file
>> python3 create_demo.py input_detections_folder output_detections_folder Shortest_Paths.txt input_frames_folder output_frames_folder output_video_path path_to_output_file .png(or .jpg depending on what format input frames are in)

Evaluating using MOT Challenge Dataset Detections:

>> python3 format_detections path_to_det.txt input_detections_folder

After which, follow steps as mentioned in previous section.

Preparing demo data:

Extract all the frames (using software like ffmpeg) and put them in input_frames_folder
Run a detector like Yolo and store the detections in input_detections_folder. Each image in input_frames_folder has a corresponding detection file with the same name as the image but ending with .txt extension. Every line inside the detection file has a detection in the following format [x_min, y_min, x_max, y_max, class]. Class is 0 for person.
Create empty output_detections_folder and output_frames_folder.
Run commands as mentioned in previous section.