search

sajjad-sh33 / YOLO_SAM2

Self-Prompting Polyp Segmentation in Colonoscopy Using Hybrid YOLO-SAM 2 Model
36 stars 5 forks source link

segment #7

Open liwenyang-911 opened 1 week ago

liwenyang-911 commented 1 week ago

Your work is excellent. Only the command of the detection example is provided in the readme file, what is the segment command?

sajjad-sh33 commented 1 week ago

Hi, To perform segmentation and calculate the results, you should use the Test.py file. This file is set up to handle the segmentation process and output the necessary results. If you have any other questions or need further details on how to use it, just let me know.

liwenyang-911 commented 1 week ago

Hi, To perform segmentation and calculate the results, you should use the Test.py file. This file is set up to handle the segmentation process and output the necessary results. If you have any other questions or need further details on how to use it, just let me know.

I executed the following command to train image segmentation: yolo segment train data=./YAML/Kvasir-seg.yaml model=./yolov8x-seg.pt epochs=50 imgsz=640 batch=0.90 name=Kvasir, but the result is very poor! Is this because the original YOLO model is used directly? What should I do if I want to use YOLO SAM 2 for training segmentation?

sajjad-sh33 commented 1 week ago

This problem happens because you are using the YOLO segmentation structure for implementing your model. In our paper, we use the YOLO detection to find boxes and then use these boxes for segmentation with the Segment Anything Model (SAM). Using the YOLO segmentation is weaker than our suggested structure, leading to poor results.

By running yolo task=detect mode=train model=yolov8l.pt imgsz=640 data=./YAML/Kvasir.yaml epochs=50 batch=0.90 name=Kvasir, you can have the pretrained YOLO detection model (also you can download them in the ./YOLO_Checkpoints repository and skip this step). With the trained model, run the Test.py to do the segmentation with the Segment Anything Model.