This is the repository for the ImageCLEF team at the NOAA-NVIDIA hackathon
Model | Status | Evaluation CLEF DATA | Evaluation NOAA DATA |
---|---|---|---|
Deep Segmentation'19 | ✅ Implemented | ✅ Test data 🔧 Val data: Implemented (numbers TBC) |
🔧 In progress |
Mask-RCNN | ✅ Implemented | 🔧 In progress | ❌ Not Implemented |
UNET | ❌ Not Implemented | ❌ Not Implemented | ❌ Not Implemented |
DETR | ❌ Not Implemented | ❌ Not Implemented | ❌ Not Implemented |
In order to obtain the training data (for the 2020 ImageCLEF competition):
wget https://annotator.uk/data/imageCLEFcoral2020_training_v4.zip
wget https://annotator.uk/data/annotations-train-NVIDIA-NOAA-2020.zip
wget https://annotator.uk/data/imageCLEFCORAL2020_GT.zip
wget https://annotator.uk/data/imageCLEFcoral2020_test_v4.zip
And unzip it using the password provided by @aCampello, for example with
unzip -j -P <password> imageCLEFcoral2020_training_v4.zip -d data/images
unzip -j -P <password> annotations-train-NVIDIA-NOAA-2020.zip -d data/images
unzip -j -P <password> imageCLEFCORAL2020_GT.zip -d data
unzip -j -P <password> imageCLEFcoral2020_test_v4.zip -d data/images_val
Clone this repository and cd imageclef-2019-code
Clone git repository for deeplabv3+:
mkdir src
cd src
git clone https://github.com/jfzhang95/pytorch-deeplab-xception.git
cd ..
Create python environment (e.g. with conda) and activate it
virtualenv -p python3.7 env/
. env/bin/activate
pip install torch torchvision cudatoolkit #For gpu
Install requirements and package
pip install -r requirements.txt
pip install -e .
First move all the CLEF images to data/images
and the csv with clef annotations in data/annotations.csv
. Then create masks with
python coralml/data/create_masks.py
And subsequently split into train and validation
python coralml/data/data_split.py
Modify the file data/instructions.json
to change hyperparameters of the network and train with:
python -m coralml --instructions data/instructions.json
python coralml/data/create_masks.py --data_folder_path data --image_folder images_val --mask_folder masks_val --annotations_file imageCLEFcoral2020_GT.csv
python coralml/ml/evaluate_clef.py --data_folder_path data --image_folder images_val --mask_folder masks_val --model_path models/test_model/model_best.pth
~~~~
conda create --name coralml python==3.6.7
source activate coralml
~~~~
---- Cuda available, Cuda version 9.0: ----
~~~~
conda install pytorch torchvision cudatoolkit=9.0 -c pytorch
~~~~
OR
---- Cuda not available: ---
~~~~
conda install pytorch-cpu torchvision-cpu -c pytorch
~~~~
pip install --upgrade pip
pip install -r requirements.txt