MOTChallenge 2015

Multiple Object Tracking Challenge ================================== ---- http://motchallenge.net ----- ---------------------------------- Version 1.0.2 This development kit provides scripts to evaluate tracking results. Please report bugs to [email protected] Requirements ============ - MATLAB - Benchmark data e.g. 2DMOT2015, available here: http://motchallenge.net/data/2D_MOT_2015/ Usage ===== To compute the evaluation for the included demo, which corresponds to the results of the CEM tracker (continuous energy minimization) on the training set of the '2015 MOT 2DMark', start MATLAB and run benchmarkDir = '../data/2DMOT2015/train/'; allMets = evaluateTracking('c2-train.txt', 'res/data/', benchmarkDir); Replace the value for benchmarkDir accordingly. You should see the following output (be patient, it may take a minute): Sequences: 'TUD-Stadtmitte' 'TUD-Campus' 'PETS09-S2L1' 'ETH-Bahnhof' 'ETH-Sunnyday' 'ETH-Pedcross2' 'ADL-Rundle-6' 'ADL-Rundle-8' 'KITTI-13' 'KITTI-17' 'Venice-2' Evaluating ... ... TUD-Stadtmitte *** 2D (Bounding Box overlap) *** Rcll Prcn FAR| GT MT PT ML| FP FN IDs FM| MOTA MOTP MOTAL 60.9 94.0 0.25| 10 5 4 1| 45 452 7 6| 56.4 65.4 56.9 ... TUD-Campus *** 2D (Bounding Box overlap) *** Rcll Prcn FAR| GT MT PT ML| FP FN IDs FM| MOTA MOTP MOTAL 58.2 94.1 0.18| 8 1 6 1| 13 150 7 7| 52.6 72.3 54.3 .................. .................. .................. ********************* Your Benchmark Results (2D) *********************** Rcll Prcn FAR| GT MT PT ML| FP FN IDs FM| MOTA MOTP MOTAL 45.3 71.7 1.30|500 81 161 258| 7129 21842 220 338| 26.8 72.4 27.4 Details ======= The evaluation script accepts 3 arguments: 1) sequence map (e.g. `c2-train.txt` contains a list of all sequences to be evaluated in a single run. These files are inside the ./seqmaps folder. 2) The folder containing the tracking results. Each one should be saved in a separate .txt file with the name of the respective sequence (see ./res/data) 3) The folder containing the benchmark sequences. The results will be shown for each individual sequence, as well as for the entire benchmark. Directory structure =================== ./res ---------- This directory contains - the tracking results for each sequence in a subfolder data - eval.txt, which shows all metrics for this demo ./utils ------- Various scripts and functions used for evaluation. ./seqmaps --------- Sequence lists for different benchmarks Version history =============== 1.0.4 - Oct 08, 2015 - Fixed bug where result has more frames than ground truth 1.0.3 - Jul 04, 2015 - Removed spurious frames from ETH-Pedcross2 result (thanks Nikos) 1.0.2 - Mar 11, 2015 - Fix to exclude small bounding boxes from ground truth - Special case of empty mapping fixed 1.0.1 - Feb 06, 2015 - Fixes in 3D evaluation (thanks Michael) 1.00 - Jan 23, 2015 - initial release 训练序列ground truth文件（txt）： 每一行表示一个目标实例，每一行有十个数。第1个数表示在第几（这个数）帧出现了目标。第2个数是一个独一无二id，表示目标属于的一个轨迹。第3—6个数代表图片中行人的bounding box的位置坐标，左上角坐标以及宽与高。第7个数是一个标志，是否这个条目会被考虑，等于0表示在评价时忽略这个特殊的实例（忽略这个目标），等于1表示标记它是active的。最后三个数都设为-1（只在3D中会用到，是世界坐标系中行人的3D位置代表人的脚）。 Benchmark Code： 文件夹res：有一个子文件夹Data，存放的是每个序列的跟踪结果。 文件夹seqmaps：说明用了哪些数据集进行训练和测试。 文件夹utils：评价函数的代码。 使用： 1. benchmarkDir = 'J:\MOT\CNN_MOTCode\scea_ver1\dataset\MOT2015\'; 给定训练数据集路径。 2. allMets = evaluateTracking('cvpr2016.txt', 'res/cvpr2016/', benchmarkDir);调用评价函数 参数：第一个参数指明用哪些数据集，第二个参数是自己跟踪算法的结果，第三个参数是数据集路径包括ground truth。