afaq-ahmad/fish-detection-and-species-classification

#################### Object Detection Steps ####################

Make a new directory for the project.

cd home/ mkdir Fish cd Fish

Training

git clone https://github.com/zylo117/Yet-Another-EfficientDet-Pytorch.git cd Yet-Another-EfficientDet-Pytorch

pip install pycocotools numpy==1.16.0 opencv-python tqdm tensorboard tensorboardX pyyaml webcolors matplotlib pip install torch==1.4.0 pip install torchvision==0.5.0 pip install efficientnet_pytorch

Make a new directory for dataset.

mkdir dataset cd dataset

Make a new directory for your_project_name and place the dataset in it.

mkdir Fish cd Fish

Place the metadata and labelled folder here.

Place the python code csv_manipulation.py and csv_to_coco.py in the dataset directory

python csv_manipulation.py python csv_to_coco.py cd ..

Adding the project file Fish.yml in projects folder

cd projects

mv Fish.yml

				project_name: Fish  # also the folder name of the dataset that under data_path folder
				train_set: train
				val_set: val
				num_gpus: 1

				# mean and std in RGB order, actually this part should remain unchanged as long as your dataset is similar to coco.
				mean: [0.10027235, 0.5572759 , 0.56590959]
				std: [0.06090582, 0.1283926 , 0.14095272]

				# this is coco anchors, change it if necessary
				anchors_scales: '[2 ** 0, 2 ** (1.0 / 3.0), 2 ** (2.0 / 3.0)]'
				anchors_ratios: '[(1.0, 1.0), (1.4, 0.7), (0.7, 1.4)]'

				# must match your dataset's category_id.
				# category_id is one_indexed,
				# for example, index of 'car' here is 2, while category_id of is 3
				obj_list: ['fish']

cd ..

Adding weights of the model

mkdir logs cd logs mkdir Fish cd Fish

mv efficientdet-d3_31_122304.pth cd ../..

Downloading pretrained weights

mkdir weights cd weights wget https://github.com/zylo117/Yet-Another-Efficient-Pytorch/releases/download/1.0/efficientdet-d3.pth cd ..

Starting the training:

-c means efficientdet-d3

python train.py -c 3 -p Fish --batch_size 16 --lr 1e-5 --num_epochs 10 --load_weights weights/efficientdet-d3.pth

Training Resume

python train.py -c 3 -p Fish --batch_size 16 --lr 1e-5 --num_epochs 10 --load_weights last

Testing on Images:

it will load the image from img_path and save in test folder:

python efficientdet_test_Fish_img.py img_path weight_loc efficientnet_version

python efficientdet_test_Fish_folder.py img_folder_path weight_loc efficientnet_version

Testing on video:

it will load the video from vid_path:

python efficientdet_test_videos.py vid_path weight_loc efficientnet_version

#################### Classification Steps ####################

Make Classification dataset. Run csv_manipulation_classification.py it will create a dataset with Classif_Dataset Folder.

python csv_manipulation_classification.py

cd ..

Make Classification Folder:

mkdir Classification mv Yet-Another-EfficientDet-Pytorch/Classif_Dataset Classification/Classif_Dataset cd Classification

Train Validation Split

python train_val_split.py

Starting the training

python main.py Classif_Dataset

Testing on Image:

python test.py --img_path=img_path --model_path=model_best.pth.tar

#################### Full Pipeline Steps ####################

Testing on image, it will save the in the same path.

python efficientdet_test_Fish.py efficientdet-d1_9_152910.pth model_best.pth.tar image_path efficientnet_version python efficientdet_test_Fish_folder.py efficientdet-d1_9_152910.pth classification.pth.tar image_path efficientnet_version

Testing on video, it will save the in the same path.

python efficientdet_test_Fish_video.py detection_efficientdet-d3.pth classification.pth.tar fish.mp4 efficientnet_version