Unable to download image files of two types:
1- the pictures
2- the objects in the images
as well as labeling objects in images as b_box_1640449327_CM TEST A4 29V CAL 25122021.labels.json
with _1640449327_CM TEST A4 29V CAL 25122021 the main image name
and for example: 1663845463_Digitization_20220922_crop_2168_0754.jpg
where 1663845463_Numérisation_20220922 is the name of the image,
and _crop_2168_0754 is the position of the crop in the image: 2168_0754 for 22 x 22 crops
With a Json in the form: b_box_1638365656_Michel_0_4_1080p.labels.json:
{
"Version 1,
“type”: “bounding box labels”,
“boundingBoxes”: {
“1646239935_CM TEST A4 10Vet IMP 20012022.jpg”: [{
“label”: “1646239935_CM TEST A4 10Vet IMP 20012022_crop_0676_3379”,
“x”:676,
“y”:3379,
“width”:22,
“height”:22
}, {
“label”: “1646239935_CM TEST A4 10Vet IMP 20012022_crop_1186_1745”,
“x”:1186,
“y”:1745,
“width”:22,
“height”:22
}, {
“label”: “1646239935_CM TEST A4 10Vet IMP 20012022_crop_1645_2162”,
“x”:1645,
“y”:2162,
“width”:22,
“height”:22
}]
}
}
Just to make sure, you are trying to upload a dataset to edge impulse correct?
Can you make sure your label file is called bounding_boxes.labels (without the .json extension).
That being said, we will release support of new image dataset acquisition formats on Monday (COCO JSON, Pascal VOC, YOLO TXT, Plain CSV, OpenImage CSV). Maybe this will help you.
FYI, we just released the support of new image dataset acquisition formats (COCO JSON, Pascal VOC, YOLO TXT, OpenImage, Plain CSV): Uploader - Edge Impulse Documentation
thanks to your example, I passed a first test pass on the target machine, but the result is not yet good. can I send you some elements to have your opinion?
a priori the objects sought are 22x11 pixels in a photo which is 2500x3500, I made an eim with 900 real objects and 15000 fake objects stretched from 22x22 to 32x32, and I am surprised that the debug obtained is in 32x32 ???
here, I changed my method to that recommended by MMarciel: complete nappies with annotations in Yolo.txt format then a Fomo treatment.
Quick question about the 1024x1024 limitation, how can I break it?
-1- by paying a license, would I have the right to make images of 2500x3500 in one piece?
-2- by cutting my 2500x3500 images into 12 833x875 images (or with 840x900 overlapping)?
-3- by resizing my 2500x3500 images into a 1024x1024 image?
Thank you for your advice
b_box_1640449327_CM TEST A4 29V CAL 25122021.labels.json
