EuroCity Persons Dataset

Back

Submission #31, by HUI_Tsinghua-Daimler Joint Research VRU3-B Project

Used Method

YOLOv3_640

Submitted on

2019-05-17 04:56:27

Description

One of the test user for the EuroCity Persons Benchmark. Trained on day-time data only. The same training set with the published ECP PAMI2019 paper, height∈[20,∞], occlusion∈[0, 80]. Based on YOLOv3 c++ version (https://github.com/pjreddie/darknet/wiki/YOLO:-Real-Time-Object-Detection ), configurations: 640*640, k-means anchor (k=9), random=1, lr = 1e-03, policy=steps 40000,45000, max_batches=50200.

Method description: One of the test user for the EuroCity Persons Benchmark. Trained on day-time data only. The same training set with the published ECP PAMI2019 paper, height∈[20,∞], occlusion∈[0, 80]. Based on YOLOv3 c++ version (https://github.com/pjreddie/darknet/wiki/YOLO:-Real-Time-Object-Detection ), configurations: 640*640, k-means anchor (k=9), random=1, lr = 1e-03, policy=steps 40000,45000, max_batches=50200.
Up-/Downsampling: 1.0
Computation Environment: Nvidia TitanX
Computation Time: 0.1

Result

day
0.273
0.564
0.623
0.456
0.252
0.544
0.596
0.435