search

dbolya / yolact

A simple, fully convolutional model for real-time instance segmentation.
MIT License
5.01k stars 1.32k forks source link

Dataset images size? #512

Open dulre opened 4 years ago

dulre commented 4 years ago

Hi i have manually labeled some images(.png and .jpg formats) with labelme and i converted the annotations to coco format using labelme2coco tool. My images are all of different size do i have to manually resize them to 512x512 ? If how i can do it without having to redo the labeling? It is fine to have pngs and jpgs in the dataset or i need to convert pngs to jpgs?

Thanks you!

suchiz commented 4 years ago

It's always resized, so you don't need to worry about labelling. Concerning the pngs, I think it would work. I didn't try but in data/coco.py where the images are loaded. 

    img = cv2.imread(path)
    height, width, _ = img.shape

It doesnt care about channels... Well not sure. Just try out

bhuvanofc commented 2 years ago

@suchiz Hi I have the same issue should I have to change my labels and my ground truth data or the network will take care of it?