This repository contains codes for the official implementation in PyTorch of UEPNet as described in Uniformity in Heterogeneity: Diving Deep into Count Interval Partition for Crowd Counting.

The codes is tested with PyTorch 1.5.0. It may not run with other versions.

The network structure of the proposed UEPNet. It consists of a simple encoderdecoder network for feature extraction and an Interleaved Prediction Head to classify each patch into certain interval.

The UEPNet achieved state-of-the-art performance on several challenging datasets with various densities, although using a quite simple network structure.

• Clone this repo into a directory named UEPNet_ROOT
• Organize your datasets as required
• Install Python dependencies. We use python 3.6.5 and pytorch 1.5.0

pip install -r requirements.txt

We use a list file to collect all the images and their ground truth annotations in a counting dataset. When your dataset is organized as recommended in the following, the format of this list file is defined as:

train/scene01/img01.jpg train/scene01/img01.txt
train/scene01/img02.jpg train/scene01/img02.txt
...
train/scene02/img01.jpg train/scene02/img01.txt

DATA_ROOT/
        |->train/
        |    |->scene01/
        |    |->scene02/
        |    |->...
        |->test/
        |    |->scene01/
        |    |->scene02/
        |    |->...
        |->train.list
        |->test.list

DATA_ROOT is your path containing the counting datasets.

For the annotations of each image, we use a single txt file which contains one annotation per line. Note that indexing for pixel values starts at 0. The expected format of each line is:

x1 y1
x2 y2
...

A trained model (with an MAE of 54.64) on SHTechPartA is available at "./ckpt", run the following commands to conduct an evaluation:

CUDA_VISIBLE_DEVICES=0 python3 test.py \
    --train_lists $DATA_ROOT/train.list \
    --test_lists $DATA_ROOT/test.list \
    --dataset_mode shtechparta \
    --checkpoints_dir ./ckpt/ \
    --dataroot $DATA_ROOT \
    --model uep \
    --phase test \
    --vgg_post_pool \
    --gpu_ids 0

• Part of codes are borrowed from the pytorch-CycleGAN-and-pix2pix.

If you find UEPNet is useful in your project, please consider citing us:

@inproceedings{wang2021uniformity,
  title={Uniformity in Heterogeneity: Diving Deep into Count Interval Partition for Crowd Counting},
  author={Wang, Changan and Song, Qingyu and Zhang, Boshen and Wang, Yabiao and Tai, Ying and Hu, Xuyi and Wang, Chengjie and Li, Jilin and Ma, Jiayi and Wu, Yang},
  journal={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  year={2021}
}

• [AAAI2021] To Choose or to Fuse? Scale Selection for Crowd Counting. (paper link & codes)
• [ICCV2021] Rethinking Counting and Localization in Crowds: A Purely Point-Based Framework. (paper link & codes)