[dev] Make masks from `x,y,w,h` as % of image

[nickvazz]

!pip install fiftyone
import fiftyone.zoo as foz
import numpy as np
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
from itertools import product

#Dwnload the COCO dataset for the "bowl" class
#Train set
dataset = foz.load_zoo_dataset(
    "coco-2017",
    split="train",
    label_types=["segmentations"],
    classes=["bowl", "teddy bear"],
    max_samples=20,
)

for idx, sample in enumerate(dataset):
    if idx > 5:
        break
    img = sample['filepath']
    img = mpimg.imread(img)
    useful_detections = [d for d in sample['ground_truth']['detections'] if d['label'] in ['teddy_bear','bowl']]
    bbox = useful_detections[0]['bounding_box']

    x,y,w,h = bbox
    print(bbox)

    mask = np.zeros(img.shape[:2])

    # print(x*img.shape[1], (x+w)*img.shape[1], y*img.shape[0], (y+h)*img.shape[0], w*img.shape[1], h*img.shape[0])
    # print(np.floor(x*img.shape[1]), np.ceil((x+w)*img.shape[1]), np.floor(y*img.shape[0]), np.ceil((y+h)*img.shape[0]), w*img.shape[1], h*img.shape[0])

    def possible_slices(x,w,y,h):
        xslices = [
                   slice(int(np.floor(x*img.shape[1])), int(np.floor((x+w)*img.shape[1]))),
                   slice(int(np.ceil(x*img.shape[1])), int(np.floor((x+w)*img.shape[1]))),
                   slice(int(np.floor(x*img.shape[1])), int(np.ceil((x+w)*img.shape[1]))),
                   slice(int(np.ceil(x*img.shape[1])), int(np.ceil((x+w)*img.shape[1]))),
        ]
        yslices = [
                   slice(int(np.floor(y*img.shape[0])), int(np.floor((y+h)*img.shape[0]))),
                   slice(int(np.ceil(y*img.shape[0])), int(np.floor((y+h)*img.shape[0]))),
                   slice(int(np.floor(y*img.shape[0])), int(np.ceil((y+h)*img.shape[0]))),
                   slice(int(np.ceil(y*img.shape[0])), int(np.ceil((y+h)*img.shape[0]))),
        ]

        return list(product(xslices,yslices))

    # xslice = slice(int(np.floor(x*img.shape[1])), int(np.floor((x+w)*img.shape[1])))
    # yslice = slice(int(np.floor(y*img.shape[0])), int(np.floor((y+h)*img.shape[0])))

    all_possible_slices = possible_slices(x,w,y,h)
    print(all_possible_slices)

    completed = False
    for xslice, yslice in all_possible_slices:
        if completed: continue
        try:
            mask[yslice, xslice] += useful_detections[0]['mask']
            completed = True
        except:
            pass

    # mask[int(y*img.shape[0]):int((y+h)*img.shape[0])+1, int(x*img.shape[1]):int((x+w)*img.shape[1])+1] += useful_detections[0]['mask']

    fig, ax = plt.subplots(ncols=3, figsize=(20,6), sharex=True, sharey=True)

    ax[0].imshow(img)
    ax[1].imshow(mask)
    ax[2].scatter(img.shape[1] * np.array([x,x,x+w,x+w]), img.shape[0] * np.array([y,y+h,y,y+h]), color='r')
    ax[2].imshow(useful_detections[0]['mask'])
    ax[0].imshow(mask, alpha=0.5,)
    ax[0].scatter(img.shape[1] * np.array([x,x,x+w,x+w]), img.shape[0] * np.array([y,y+h,y,y+h]), color='r')
    plt.tight_layout()
    plt.show()

[nickvazz]

https://diyago.github.io/2019/08/25/road-detection.html

[nickvazz]

https://blog.ceshine.net/post/custom-image-augmentation-with-keras/

[nickvazz]

https://albumentations.ai/docs/external_resources/blog_posts_podcasts_talks/#blog-posts

[nickvazz]

Add all of the useful detections

for idx, sample in enumerate(dataset):
    if idx > 5:
        break
    img = sample['filepath']
    img = mpimg.imread(img)
    useful_detections = [d for d in sample['ground_truth']['detections'] if d['label'] in ['teddy_bear','bowl']]
    print(bbox)

    mask = np.zeros(img.shape[:2])

    # print(x*img.shape[1], (x+w)*img.shape[1], y*img.shape[0], (y+h)*img.shape[0], w*img.shape[1], h*img.shape[0])
    # print(np.floor(x*img.shape[1]), np.ceil((x+w)*img.shape[1]), np.floor(y*img.shape[0]), np.ceil((y+h)*img.shape[0]), w*img.shape[1], h*img.shape[0])

    def possible_slices(x,w,y,h):
        xslices = [
                   slice(int(np.floor(x*img.shape[1])), int(np.floor((x+w)*img.shape[1]))),
                   slice(int(np.ceil(x*img.shape[1])), int(np.floor((x+w)*img.shape[1]))),
                   slice(int(np.floor(x*img.shape[1])), int(np.ceil((x+w)*img.shape[1]))),
                   slice(int(np.ceil(x*img.shape[1])), int(np.ceil((x+w)*img.shape[1]))),
        ]
        yslices = [
                   slice(int(np.floor(y*img.shape[0])), int(np.floor((y+h)*img.shape[0]))),
                   slice(int(np.ceil(y*img.shape[0])), int(np.floor((y+h)*img.shape[0]))),
                   slice(int(np.floor(y*img.shape[0])), int(np.ceil((y+h)*img.shape[0]))),
                   slice(int(np.ceil(y*img.shape[0])), int(np.ceil((y+h)*img.shape[0]))),
        ]

        return list(product(xslices,yslices))

    for detection in range(len(useful_detections)):
        bbox = useful_detections[detection]['bounding_box']
        x,y,w,h = bbox
        all_possible_slices = possible_slices(x,w,y,h)

        completed = False
        for xslice, yslice in all_possible_slices:
            if completed: continue
            try:
                mask[yslice, xslice] += useful_detections[detection]['mask']
                completed = True
            except:
                pass

    # mask[int(y*img.shape[0]):int((y+h)*img.shape[0])+1, int(x*img.shape[1]):int((x+w)*img.shape[1])+1] += useful_detections[0]['mask']

    fig, ax = plt.subplots(ncols=3, figsize=(20,6), sharex=True, sharey=True)
    fig.suptitle(len(useful_detections))
    ax[0].imshow(img)
    ax[1].imshow(mask)
    ax[2].scatter(img.shape[1] * np.array([x,x,x+w,x+w]), img.shape[0] * np.array([y,y+h,y,y+h]), color='r')
    ax[2].imshow(useful_detections[0]['mask'])
    ax[0].imshow(mask, alpha=0.5,)
    ax[0].scatter(img.shape[1] * np.array([x,x,x+w,x+w]), img.shape[0] * np.array([y,y+h,y,y+h]), color='r')
    plt.tight_layout()
    plt.show()