Introduction

This is a project for garbage dump detection with BCA-Net, which can be used to perform global garbage dump detection with our upcoming multi-category garbage dump dataset.

Prerequisites

• OS: Ubuntu 16.04
• GPU: Nvidia GTX/RTX series GPU with proper NVIDIA Driver installed
• Software: Docker installed

Installation

1. Create a new project folder.

   mkdir /home/$[YOUR_USERNAME]/garbage_dump
   cd /home/$[YOUR_USERNAME]/garbage_dump

2. Download the code.

   git clone https://github.com/DongshuoYin/garbage_dump_detection.git

3. Download dataset in our paper's link.

4. Unzip the dataset to ./garbage_dump_detection/data/.

   garbage_dump_detection
   ├── mmdet
   ├── tools
   ├── configs
   ├── data
   │   ├── garbage_dump_2022
   │   │   ├── VOC2012
   │   │   │   ├──train
   │   │   │   │   ├──Annotations
   │   │   │   │   ├──JPEGImages
   │   │   │   │   ├──train.txt
   │   │   │   ├──test
   │   │   │   │   ├──Annotations
   │   │   │   │   ├──JPEGImages
   │   │   │   │   ├──test.txt
   ......
   ......
   

5. Get the docker image from Docker-hub.

   sudo docker pull y389164605/garbage_dump_detection:latest

6. Create a docker container with the above image.

   sudo nvidia-docker run --privileged=true --name=$[YOUR_CONTAINER_NAME] --shm-size=8g -d -p $[PORT_FOR_CONTAINER_PORT_22]:22 -v /home/$[YOUR_USERNAME]/garbage_dump/garbage_dump_detection:/garbage_dump_detection y389164605/garbage_dump_detection:latest /usr/sbin/sshd -D

   Note:

   a. If the terminal remains inactive, create a new terminal and continue the operation.

   b. /home/$[YOUR_USERNAME]/garbage_dump/garbage_dump_detection in your computer and /garbage_dump_detection in your docker container are a pair of mapped folders and they will remain consistent.

7. Enter the above docker container.

   sudo docker exec -it $[YOUR_CONTAINER_NAME] /bin/bash
   cd /garbage_dump_detection
   python setup.py develop

PS: Installation can be completed in about 0.5~1 hour with good internet access.

Demo

1. Download the pre-trained model here and put it in /home/$[YOUR_USERNAME]/garbage_dump/garbage_dump_detection/checkpoint_backup/

2. Run the following code in container.

   cd tools
   python demo.py

PS: Demo can be completed in less than 20 seconds.

Batch inference

1. Download the pre-trained model here (same as Demo) and put it in /home/$[YOUR_USERNAME]/garbage_dump/garbage_dump_detection/checkpoint_backup/.

2. Resize your images to 1024*1024 pixels.

3. Copy all your images to /home/$[YOUR_USERNAME]/garbage_dump/garbage_dump_detection/tools/batch_inference_data/.

4. Run the following code in container.

   cd tools
   python inference.py

5. Check your imference results in /home/$[YOUR_USERNAME]/garbage_dump/garbage_dump_detection/tools/batch_inference_data/inference_visualization

Training the BCA-Net

1. Run the following code.

   cd tools 
   python train.py ../_myconfigs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712_all_layer_SE_with_ClassBalancedDataset_and_low_nms_score_config_and_data_augumentation.py

2. Check the evaluation metric on test set after each epoch.

Note: If you want to train with your own dataset, replace the dataset in /home/$[YOUR_USERNAME]/garbage_dump/garbage_dump_detection/data with yours and keep the data and folder format the same as ours.