This repository is for a paper on a method to segment petals from CT images of C. japonica.
This project can be set up in two ways: by installing the necessary packages directly on your system or by using Docker. Below are the instructions for both methods.
Please refer to get_started.md for installation and dataset preparation.
Alternatively, you can build the environment using Docker. This method ensures that the project runs in a consistent and isolated environment, regardless of your local setup. You will need Docker installed on your system. If you don't have Docker, follow the official Docker installation guide.
Once Docker is installed, you can build the Docker image for this project using the provided Dockerfile:
# Build the Docker image from the Dockerfile
docker build -t mmdetection docker/After building the image, you can run the project inside a Docker container:
# Run the project in a Docker container
docker run --gpus all --shm-size=8g -it -v {DATA_DIR}:/mmdetection/petal_ct_crop_seg/data mmdetectionThe -v {DATA_DIR}:/mmdetection/petal_ct_crop_seg/data can be optionally configured.
This command sets up the environment as specified in the Dockerfile, including the correct version of mmdetection (v2.28.2) and any other dependencies, ensuring a consistent setup across different machines.
Before you begin, make sure you navigate to the mmdetection directory. This is where you'll perform all subsequent commands related to this project.
cd path/to/mmdetectionReplace path/to/mmdetection with the actual path to the mmdetection directory on your system. This will change your current working directory to mmdetection, ensuring that all commands you run next are executed in the correct directory context.
# Train with a single GPU
python tools/train.py <CONFIG_FILE> [--gpu-id <GPU_ID>] [optional arguments]
# Train with multiple GPUs
tools/dist_train.sh <CONFIG_FILE> <GPU_NUM> For example, to train a Hybrid Task Cascade model with a ResNeXt101-64x4d backbone and 4 gpus, run:
tools/dist_train.sh /mmdetection/petal_ct_crop_seg/configs/htc_x101_64x4d_fpn_16x1_crop-rotate_50e_adam_coco_segm.py 4# single-gpu testing (w/o bbox result)
python tools/test.py <CONFIG_FILE> <DET_CHECKPOINT_FILE> --eval segm
# multi-gpu testing (w/o bbox result)
tools/dist_test.sh <CONFIG_FILE> <DET_CHECKPOINT_FILE> <GPU_NUM> --eval segmThis code segments petals in cropped images using a trained model.
python crop_seg_mmdet.py [-h] [--device DEVICE] [--input INPUT] [--output OUTPUT] config checkpoint start end
# positional arguments:
# config train config file path
# checkpoint checkpoint file
# start starting image number
# end ending image number
# options:
# -h, --help show this help message and exit
# --device DEVICE GPU device number. Dafault is 0.
# --input INPUT Directory path where images are stored. Default is /mmdetection/petal_ct_crop_seg/data/volume_1
# --output OUTPUT Directory path to output results. Default is /mmdetection/petal_ct_crop_seg/crop_ct_segThis code integrates the segmentation results of the cropped image.
python 2d_inte.py [-h] [--input INPUT] [--output OUTPUT] [--proc-num PROC_NUM] start end
# positional arguments:
# start starting image number
# end ending image number
# options:
# -h, --help show this help message and exit
# --input INPUT Directory containing the segmentation results of the cropped image. Default is /mmdetection/petal_ct_crop_seg/data/crop_ct_seg
# --output OUTPUT Directory path to output results. Default is /mmdetection/petal_ct_crop_seg/data/2d_inte/
# --proc-num PROC_NUM Number of cores used. Default is the maximum number of cores that can be used.This code integrates segmentation results that are integrated in 2D in a 3D direction.
python 3d_inte.py [-h] [--input INPUT] [--output OUTPUT]
# options:
# -h, --help show this help message and exit
# --input INPUT Directory where the segmentation results merged in 2D are stored. Default is /mmdetection/petal_ct_crop_seg/data/2d_inte/
# --output OUTPUT Directory path to output results. Default is /mmdetection/petal_ct_crop_seg/data/3d_inte/This project has been developed and tested in the following software and hardware environment.
- PyTorch Version: 1.13.1
- CUDA Version: 11.6
- cuDNN Version: 8
- Python Version: 3.10.8
- mmdetection Version: 2.28.2
-
GPU Server:
- GPU: NVIDIA TITAN RTX with 24 GB of memory
-
CPU Server:
- Processor: Intel Xeon Gold 5118
- Memory: 128 GB of RAM
Please ensure that your environment matches or is compatible with these specifications to replicate our results and effectively utilize the codebase.
We list some documents and tutorials from MMDetection, which may be helpful to you.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent