Skip to content

Latest commit

 

History

History

head_detection

Head Detection

We use the brainwash dataset introduced by paper End-to-end people detection in crowded scenes.

Recent Update

  • 2019.09.23 model v1 for brainwash dataset is released.
  • 2019.09.26 brainwash dataset (and packed pkl) is uploaded for downloading.

Brief Introduction to Model Version

  • v1 - is designed for brainwash dataset, covering head scale [10, 160]. It has 4 branches. Please check ./symbol_farm/symbol_structures.xlsx for details.

Inference Latency

  • Platform info: NVIDIA RTX 2080TI, CUDA 10.0, CUDNN 7.4.2, TensorRT 5.1.5.0
Model Version 320×240 640×480 1280×720 1920×1080 3840×2160 7680×4320
v1 0.83ms(1198.38FPS) 1.91ms(524.14FPS) 4.83ms(206.92FPS) 10.62ms(94.19FPS) 42.28ms(23.65FPS) 166.81ms(5.99FPS)
  • Platform info: NVIDIA GTX 1060(laptop), CUDA 10.0, CUDNN 7.4.2, TensorRT 5.1.5.0
Model Version 320×240 640×480 1280×720 1920×1080 3840×2160
v1 1.62ms(618.53FPS) 4.83ms(207.06FPS) 13.67ms(73.18FPS) 30.01ms(33.32FPS) 121.15ms(8.25FPS)

CAUTION: The latency may vary even in the same setting.

Accuracy on Brainwash Dataset

We train v1 on the training set (10769 images with 81975 annotated heads) and evaluate on the test set (500 images with 5007 annotated heads). This dataset is relatively simple due to monotonous scenario.

Quantitative Results on Test Set

Average Precision (AP) is used for measuring the accuracy. In detail, we use code Object-Detection-Metrics for calculating the AP metric. The following table presents the results:

Method AP
ReInspect, Lhungarian [1] 0.78
FCHD [2] 0.70
v1 (our) 0.91

[1] End-to-end people detection in crowded scenes

[2] FCHD: Fast and accurate head detection in crowded scenes

The v1 significantly outperforms the existing methods.

Some Qualitative Results on Test Set

image image image image

User Instructions

Please refer to README in face_detection for details.

Data Download

We provide original and packed data of brainwash dataset. We provide three ways to download the packed data: