[T-PAMI] NWPU-Crowd: A Large-Scale Benchmark for Crowd Counting and Localization

Update

  • [2020.08.26] Online evaluation for crowd localization is opened.
  • [2020.08.03] Annotation tool for crowd counting and localization is released.
  • [2020.07.30] Version 3 (Final Version) of the paper is pre-printed.
  • [2020.07.29] Code for crowd localization is released.
  • [2020.05.25] Version 2 of the paper is pre-printed.
  • [2020.05.22] Box-level annotations are provided. jsons.zip and mats.zip are changed. Please refer to Download section.
  • [2020.01.10] NWPU-Crowd and CrowdBenchmark for counting are released.

Overview

NWPU consists of 5,109 images and contains 2,133,375 annotated instances with point and box lables. Compareing with the previous datasets, the main advantages are:

  • Negative Samples
  • Fair Evaluation
  • High Resolution
  • Large Appearance Variation

Evaluation Protocol

Crowd Counting:

  • MAE (Overall MAE is primary key.)
  • MSE
  • NAE

Crowd Localization:

  • Precision
  • Recall
  • F1-Measure (Overall F1-Measure is primary key.)

Leaderboard

Crowd Counting: Link to CorwdBenchmark

Crowd Localization: Link to CorwdBenchmark

Download

  • Paper:
    • [Version 3] NWPU-Crowd: A Large-Scale Benchmark for Crowd Counting and Localization
    • [Version 2] NWPU-Crowd: A Large-Scale Benchmark for Crowd Counting and Localization
    • [Version 1] NWPU-Crowd: A Large-Scale Benchmark for Crowd Counting
  • Dataset: [OneDrive] [BaiduNetdisk] [CorwdBenchmark]
  • Open-source Code: [Counting Code] [Localization Code] [Annotation Tools]
  • Pre-trained Models [Link]: MCNN, VGG, CSR, CAN, SCAR, SFCN+, etc.
  • Validation Reults
    • Visualization Results for Density Prediction: [Link]: MCNN, PCC, Reg+Det, VGG, CSR, CAN, SCAR, SFCN+, etc.
    • Results for Localization: [Link]: Faster RCNN, TinyFaces, VGG+GPR, RAZ_loc.

Submission

In addition to submit the test results in CorwdBenchmark.com, you also send your validation results (density map or localization results) to us (gjy3035 [at] gmail [dot] com). We will upload it to netdisk and share its download link, which will make it easier for researchers to compare algorithms.

Some Visualization Results

  • Crowd Counting
  • Crowd Localization

Citation

If you find our project is useful for your research, please cite:

@article{gao2020nwpu,
title={NWPU-Crowd: A Large-Scale Benchmark for Crowd Counting and Localization},
author={Wang, Qi and Gao, Junyu and Lin, Wei and Li, Xuelong},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
doi={10.1109/TPAMI.2020.3013269},
year={2020}
}