LFEF

Performance

SYN70K ( $256 \times 256$ )

Model	Size	Param.	FLOPs	DS01	DS02	DS03
LFEF	0.853MB	0.223M	0.985G	73.07%	73.04%	73.42%

# Example
python val.py -ti /home/yaocong/Experimental/Dataset/SYN70K_dataset/testing_data/DS01/images/ -tm /home/yaocong/Experimental/Dataset/SYN70K_dataset/testing_data/DS01/masks/ -m /home/yaocong/Experimental/speed_smoke_segmentation/trained_models/best.pth

You will get the results:

model_name: <module 'models.LFEF' from '/home/yaocong/Experimental/speed_smoke_segmentation/models/LFEF.py'>
Testing on device cuda.
model size:                     0.850 MB
Computational complexity(FLOPs):   983.66 MMac
Number of parameters:           222.89 k
model path: ./trained_models/best.pth
test_data: /home/yaocong/Experimental/Dataset/SYN70K_dataset/testing_data/DS01/images/
Number of test Images: 1000
100%|█| 1000/1000 [00:07<00:00, 135.11it/s, test_hd=12.3, test_loss=0.328, test_
FPS:135.1
totally cost: 0m 7s
loss: 0.3280
mIoU: 73.14%

Training

Single GPU training

# Example
python main.py -bs 32 -train_dataset Host_SYN70K -e 500 -wn base

Multiple GPU training

# Example
python main.py -bs 32 -train_dataset Host_SYN70K -e 500 -wn base -gpus 0,1

Parameter	Description	Default Value
-train_dataset --train_dataset_path	Path to the training dataset.	Host_SYN70K
-validation_dataset --validation_dataset_path	Path to the validation dataset.	Host_DS0123
-ti --train_images	Path to the directory containing training images.	None
-tm --train_masks	Path to the directory containing training masks.	None
-vi --validation_images	Path to the directory containing validation images.	None
-vm --validation_masks	Path to the directory containing validation masks.	None
-bs --batch_size	Batch size for training.	8
-nw --num_workers	Number of workers for data loading.	1
-e --epochs	Number of epochs for training.	150
-lr --learning_rate	Learning rate for the optimizer.	0.001
-wd --weight_decay	Weight decay for the optimizer.	0.00001
-savedir --model_save_dir	Directory to save the trained models.	./trained_models/
-device	Device to run the training on. Choose between 'CPU' and 'GPU'.	GPU
-gpus	GPU devices to use for training. For multiple GPUs, separate by comma.	0
-resume	Path to the last checkpoint. Use this to resume training.	None
-wn --wandb_name	Name of the Weights & Biases run. Use 'no' to disable Weights & Biases.	no
-wid --wandb_id	Weights & Biases run ID.	None
-sti --save_train_image	Save the training images. Include this argument to enable this feature.	False
-svil --save_validation_image_last	Save the last validation image. Include this argument to enable this feature.	False
-svib --save_validation_image_best	Save the best validation image. Include this argument to enable this feature.	False

Inference

val

# Example
python val.py -ti /home/yaocong/Experimental/Dataset/SYN70K_dataset/testing_data/DS01/images/ -tm /home/yaocong/Experimental/Dataset/SYN70K_dataset/testing_data/DS01/masks/ -m /home/yaocong/Experimental/speed_smoke_segmentation/trained_models/last.pth

Parameter	Description	Default Value
-ti, --test_images	Path to the directory containing test images.	None (required)
-tm, --test_masks	Path to the directory containing test masks.	None (required)
-bs, --batch_size	Batch size for testing.	1
-nw, --num_workers	Number of workers for data loading during testing.	1
-m, --model_path	Path to the trained model to be used for testing.	"./trained_models/best.pth"
-wn, --wandb_name	Name of the Weights & Biases run for testing. Use 'no' to disable Weights & Biases.	"no"

Analytical model

# Example
python inference_multiple_pictures_for_evaluate.py -td /home/yaocong/Experimental/Dataset/SYN70K_dataset/testing_data/DS01/ -m /home/yaocong/Experimental/speed_smoke_segmentation/trained_models/best.pth

Parameter	Description	Default Value
-td, --test_directory	Path to the directory containing test images.	None (required)
-m, --model_path	Path to the trained model to be used for evaluation.	None (required)

Demo

# Example
python smoke_segmentation.py -s /home/yaocong/Experimental/Dataset/smoke_video_dataset/Black_smoke_517.avi -m /home/yaocong/Experimental/speed_smoke_segmentation/trained_models/best.pth -show

Parameter	Description	Default Value
-s, --source	Path to the image, video file, or directory to be tested.	None (required)
-m, --model_path	Path to the trained model to be used for smoke segmentation.	None (required)
-vtf, --video_to_frames	Convert the video to frames. Include this argument to enable this feature.	False
-save, --save_video	Save the output video. Include this argument to enable this feature.	False
-show, --show_video	Display the output video. Include this argument to enable this feature.	False

Citation

Acknowledgements