B2Q-Net 🚀

This repository contains the code for B2Q-Net: Bidirectional Branch Query Network for Surgical Phase Recognition.

Overview

🏆 Achievements

We're thrilled to announce that B2Q-Net secured 2nd place in the prestigious 2025 APTOS Big Data Competition! 🎉

Check out the Competition Ranking to see our achievement! 🔝

Check out the solution to learn about our specific solution! 💡

⚙️ Installation

The project was modified from BNpitfalls, thanks to their wonderful work!

💻 Recommended Environment:

• Python 3.9
• Cuda 11.6
• PyTorch 1.12.0

To install the required dependencies, simply run:

pip install -r requirements.txt

📦 Data Preparation

Step 1:

Download Cholec80, M2CAI16 and AutoLaparo datasets

• Access can be requested Cholec80, M2CAI16, AutoLaparo.
• Download the videos for each datasets and extract frames at 1fps. E.g. for video01.mp4 with ffmpeg, run:

mkdir /<PATH_TO_THIS_FOLDER>/data/frames_1fps/01/
ffmpeg -hide_banner -i /<PATH_TO_VIDEOS>/video01.mp4 -r 1 -start_number 0 /<PATH_TO_THIS_FOLDER>/data/frames_1fps/01/%08d.jpg

• The final dataset structure should look like this:

B2Q-Net/
	data/
		frames_1fps/
			01/
				00000001.jpg
				00000002.jpg
				00000003.jpg
				00000004.jpg
				...
			02/
				...
			...
			80/
				...
		phase_annotations/
			video01-phase.txt
			video02-phase.txt
			...
			video80-phase.txt
		tool_annotations/
			video01-tool.txt
			video02-tool.txt
			...
			video80-tool.txt
	output/
	train_scripts/

Step 2:

Download pretrained model (ConvNeXt V2-T)

• download ConvNeXt V2-T weights and place here: train_scripts/convnext/convnextv2_tiny_1k_224_ema.pt

Training and Evaluation

• Train our B2Q-Net on the Cholec80 dataset as an example. This will create an experiment folder under ./output to store training configurations, logs, and checkpoints.

python3 train.py phase --split cuhk4040 --bn_off --backbone convnextv2 --workers 4 --freeze --seq_len 256 --lr 1e-4 --random_seed --trial_name TRIAL_NAME --cfg configs/config.yaml

• Evaluate the trained model.

python3 save_predictions.py phase --split cuhk4040 --backbone convnextv2 --seq_len 256 --resume ../output/checkpoints/phase/[TRIAL_NAME]/models/checkpoint_best_acc.pth.tar --cfg configs/config.yaml

Contact

Wenjie Zhang (zwjie@mail.sdu.edu.cn)