Bubble Counting with FamNet

This repo introduces a method for implementing the FamNet on a large image set where it is inconvenient to manually select exemplars for every images. It also contains the experiemnts conducted for investigating the quality of the counting result. FamNet is based from this paper:

Learning To Count Everything
Viresh Ranjan, Udbhav Sharma, Thu Nguyen and Minh Hoai
Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.

Link to arxiv preprint: https://arxiv.org/pdf/2104.08391.pdf

Introduction

The notebooks, "EXP_9_Batches" and "EXP_1_Batch", shows the experiments of the counting pipeline. The pipeline modules are saved in the package BubbleCount. The directories Exemplars, Targets, and Outputs are used in the pipeline

Visualization

Set up by creating a virtual environment and installing requirements.

python -m venv .venv
source .venv/Scripts/Activate
pip install -r requirements.txt

Run Visualize Counts.ipynb and modify the following in args to your liking:

"result_path": "./2025/Outputs/out.csv" Directory to store prediction CSV.

"output_dir": "./2025/Outputs/" Directory for predicted images.

"raw_img_dir": "./2025/Images/SEN10_1.8_6_300-400" Raw images to be cropped.

"target_path": "./2025/Targets" Cropped images to be predicted.

Uncomment the following line in the 4th cell to crop raw images (only needs to be run once): image_preprocess.crop_to_interest(image_dir=args["raw_img_dir"],region=[65, 770, 1090, 970],output_dir=args["target_path"])