Crowd Counting With Partial Annotations in an Image

This repo is the official implementation of paper "Crowd Counting With Partial Annotations in an Image" (ICCV2021).

Usage

Generate Random Masks

Firstly, modify the dataset root and change the root on your disk in line 13 on 'data_prapare/make_dataset_generate_masks.py':

raw_data_root = '/p300/2020-Counting/datasets/raw/' + dataset_name + '/train_data/'

For example, to generate the random picked masks 10% on Part A, where the raw_data_root is '/p300/2020-Counting/datasets/raw/part_A_final/train_data/' run:

python data_prapare/make_dataset_generate_masks.py

It will generate the random 10% mask on the root '/p300/2020-Counting/datasets/masked_data/part_A_final/train_data/2020_12_16_v2_rand_mask/'.

We also upload the generated masks in our experiment in Google Drive and here are some examples.