# train efficientdet-d2 on a custom dataset with pretrained weights
# with batchsize 8 and learning rate 1e-5 for 10 epoches

python train.py -c 2 -p your_project_name --batch_size 8 --lr 1e-5 --num_epochs 10 \
 --load_weights /path/to/your/weights/efficientdet-d2.pth

# with a coco-pretrained, you can even freeze the backbone and train heads only
# to speed up training and help convergence.

python train.py -c 2 -p your_project_name --batch_size 8 --lr 1e-5 --num_epochs 10 \
 --load_weights /path/to/your/weights/efficientdet-d2.pth \
 --head_only True

# while training, press Ctrl+c, the program will catch KeyboardInterrupt
# and stop training, save current checkpoint.

# let says you started a training session like this.

python train.py -c 2 -p your_project_name --batch_size 8 --lr 1e-5 \
 --load_weights /path/to/your/weights/efficientdet-d2.pth \
 --head_only True
 
# then you stopped it with a Ctrl+c, it exited with a checkpoint

# now you want to resume training from the last checkpoint
# simply set load_weights to 'last'

python train.py -c 2 -p your_project_name --batch_size 8 --lr 1e-5 \
 --load_weights last \
 --head_only True

# use tensorboard to look at the train and val plots

tensorboard --logdir logs/your_project_name/tensorboard/ --port 6006

# when you get bad result, you need to debug the training result.
python train.py -c 2 -p your_project_name --batch_size 8 --lr 1e-5 --debug True

# then checkout test/ folder, there you can visualize the predicted boxes during training
# don't panic if you see countless of error boxes, it happens when the training is at early stage.
# But if you still can't see a normal box after several epoches, not even one in all image,
# then it's possible that either the anchors config is inappropriate or the ground truth is corrupted.

# you can train model with docker
#first build image
docker build -f Dockerfile.train -t tsd/train .

#input, output and weights folders should be mounted as volume
	#-v C:\dataset:/app/input/ \    - folder with datasets
	#-v C:\output:/app/output/ \    - folder for logs and checkpoint weights
	#-v C:\weights:/app/weights/    - folder with weights and project files
##then run with arguments in yaml project file
  --compound_coef COMPOUND_COEF
                        coefficients of efficientdet
  --num_workers NUM_WORKERS
                        num_workers of dataloader
  --batch_size BATCH_SIZE
                        The number of images per batch among all devices
  --head_only HEAD_ONLY
                        whether finetunes only the regressor and the
                        classifier, useful in early stage convergence or
                        small/easy dataset
  --lr LR
  --optim OPTIM         select optimizer for training, suggest using 'admaw'
                        until the very final stage then switch to 'sgd'
  --num_epochs NUM_EPOCHS
  --val_interval VAL_INTERVAL
                        Number of epoches between valing phases
  --save_interval SAVE_INTERVAL
                        Number of steps between saving
  --es_min_delta ES_MIN_DELTA
                        Early stopping's parameter: minimum change loss to
                        qualify as an improvement
  --es_patience ES_PATIENCE
                        Early stopping's parameter: number of epochs with no
                        improvement after which training will be stopped. Set
                        to 0 to disable this technique.
  --load_weights LOAD_WEIGHTS
                        whether to load weights from a checkpoint, set None to
                        initialize, set 'last' to load last checkpoint
  --batch_multiplier BATCH_MULTIPLIER
                        Increasing Mini-batch Size without Increasing Memory
  
#example
docker run --gpus=all -it --rm -v C:\EfficientDet-Pytorch\datasets\:/app/input/ \
-v C:\EfficientDet-Pytorch\logs:/app/output/ \
-v C:\EfficientDet-Pytorch\logs\synth_signs:/app/weights/ \
tsd/train 
#project.yml
	compound_coef: 3,
	load_weights: best_mapillary_efficientdet-d3_0_1.pth,
	batch_size: 1, 
	num_workers: 0,
	head_only: True

# eval on your_project, efficientdet-d3

python coco_eval.py -p your_project_name -c 3 \
 -w /path/to/your/weights

#detect traffic signs on a video

python inference.py -p your_project_name -c 3 -i path/to/video/or/folder -o path/to/output \
 -w /path/to/your/weights

#visualize bboxes
python visualize_inference.py -p your_project_name -c 3 --video_path path/to/video/ --detection_path path/to/predicted/file \
 --destination path/to/output.avi

#put class image instead of the classname
python visualize_inference.py -p your_project_name -c 3 --video_path path/to/video/ --detection_path path/to/predicted/file \
 --destination path/to/output.avi --template-dir path/to/folder/with/images

# you can use inference with docker
#first build image
docker build -f Dockerfile.inference -t tsd/inference .

#input, output and weights folders should be mounted as volume
	#-v C:\dataset:/app/input/ \    - contains videos for detection
	#-v C:\output:/app/output/ \    - folder for json prediction
	#-v C:\weights:/app/weights/    - folder with weights and project files
#then run with arguments in yaml project file
  --compound_coef COMPOUND_COEF
                        coefficients of efficientdet
  --weights WEIGHTS
                        /path/to/weights
  --nms_threshold NMS_THRESHOLD
                        nms threshold, don't change it if not for testing
                        purposes
  --threshold THRESHOLD
                        threshold of classification, don't change it if not
                        for testing purposes
  --cuda CUDA
  --device DEVICE
  --float16 FLOAT16
  --batch_size BATCH_SIZE
                        The number of images per batch among all devices
  --num_workers NUM_WORKERS
                        num_workers of dataloader
  --tracking TRACKING   Apply tracker
  --max_age MAX_AGE     Max frames to keep object id without bbox
  --min_hits MIN_HITS   Min frames with object set id
 
#example
docker run --gpus=all -it --rm -v C:\EfficientDet-Pytorch\input\:/app/input/ \
-v C:\EfficientDet-Pytorch\output:/app/output/ \
-v C:\EfficientDet-Pytorch\logs\synth_signs:/app/weights/ \
tsd/inference 
#project.yml
	compound_coef: 3,
	weights: "best_mapillary_efficientdet-d3_0_1.pth",
	batch_size: 1,
	num_workers: 0