Fastai Image crop

neumann · June 14, 2019, 2:42pm

how can i crop a fastai image by the coordinates of a bounding box ?

I am detection objects, but would like to crop the images and save them.

thanks,

muellerzr · June 14, 2019, 2:44pm

I imagine you could utilize the PIL library to do this. I did this for YOLOv3 when I wanted to do the exact same thing. If you are given the coordinates you can go into PIL and save the new image based on the crop.

neumann · June 14, 2019, 2:49pm

can you share a code snippet, i am getting a little bit confused by what is what in fastai

I have this image object i would like to crop

img.size
torch.Size([512, 512])
type(img)
fastai.vision.image.Image

muellerzr · June 14, 2019, 2:51pm

How are you getting the coordinates? Via learn.predict()? Or where are they coming from

neumann · June 14, 2019, 2:52pm

yes exactly, i’m following these examples here:

github.com

ChristianMarzahl/ObjectDetection/blob/master/helper/object_detection_helper.py#L172




def show_preds(img, bbox_pred, preds, scores, classes, figsize=(5,5)):


    _, ax = plt.subplots(1, 1, figsize=figsize)
    for bbox, c, scr in zip(bbox_pred, preds, scores):
        img.show(ax=ax)
        txt = str(c.item()) if classes is None else classes[c.item()+1]
        draw_rect(ax, [bbox[1],bbox[0],bbox[3],bbox[2]], text=f'{txt} {scr:.2f}')




def show_results_side_by_side(learn: Learner, anchors, detect_thresh:float=0.2, nms_thresh: float=0.3,  image_count: int=5):


    with torch.no_grad():
        img_batch, target_batch = learn.data.one_batch(DatasetType.Valid, False, False, False)


        prediction_batch = learn.model(img_batch[:image_count])
        class_pred_batch, bbox_pred_batch = prediction_batch[:2]


        bbox_gt_batch, class_gt_batch = target_batch[0][:image_count], target_batch[1][:image_count]


        for img, bbox_gt, class_gt, clas_pred, bbox_pred in list(

muellerzr · June 14, 2019, 2:54pm

Ah I see now!
Okay say we have the coordinates as top-left followed by bottom-right. You could do the following:
img = Image.open("example.png")
img2 = img.crop((x1,y1,x2,y2))

Does this help?

neumann · June 14, 2019, 2:57pm

i wanted to do it on the fastai.Image object , is that possible, do i have to go back and reopen the image using PIL ?

muellerzr · June 14, 2019, 2:57pm

The fastai image object should be a PIL image if I am not mistaken.

Edit: it is. So you should be able to repeat the above with it. Let me know if you cannot

neumann · June 14, 2019, 3:00pm

i wasn’t able to repeat the above, i am stuck on this for a few hours now

it might be two reasons img.crop() might be using the fastai crop() which is different from the PIL crop or i have the coordinates mixed up, i get confused when its center,w,h when its top left, bottom right, etc…

muellerzr · June 14, 2019, 3:02pm

Ah wait! On your image. do im. followed by a tab and look through the options. I believe one of them should be the actual image, it converts it to a tensor. I can verify this when I am at a computer next.

neumann · June 14, 2019, 3:03pm

img.data or img.px i guess.

no worries, thanks for your help, will let you know if get somewhere with this.

neumann · June 14, 2019, 3:08pm

got it !!!

happy days !

print(img.px.shape)

img.px = img.px[:, 295:295+77,262:252+94]

print(img.px.shape)

torch.Size([3, 512, 512])
torch.Size([3, 77, 84])

Krieker · October 10, 2019, 12:07pm

How do you convert predicted bounding box (which has values [-1;1]) into these numbers [:, 295:295+77,262:252+94]?

Adleman · March 25, 2021, 7:48am

How did you get the predicted values?