Preparing Custom Dataset

[anmspro]

I have a dataset folder which contains three folders named images, tags, masks. Each of these folders contains images. For example:
In images:

In tags:

In masks:

So, I have the bounding box region cropped and a mask from that cropped region. All these are in image format.

How do I prepare this dataset to use for training on a pretrained Mask-RCNN model?

[konstantin-frolov]

See information about pycococreator.

[anmspro]

See information about pycococreator.

Can you please elaborate?

[konstantin-frolov]

Can you please elaborate?

Create your own COCO-style datasets

[anmspro]

Can you please elaborate?

Create your own COCO-style datasets

Hey, man!
I have read the blog. I tried to use the code with some modification. Still couldn't generate dataset.
I got an error while trying it. Could you please take a look and provide some guidelines?
Here's the link - waspinator/pycococreator#34

[gethubwy]

See information about pycococreator.

Hello, Excuse me, I want to know how to train a new mode, I use the software named "labelme" to label my data, I want to train a new model starting from pre-trained COCO weights, but I don't know how to write the annotation file in validate set, are there tools to help me? Thank you.

[konstantin-frolov]

See information about pycococreator.

Hello, Excuse me, I want to know how to train a new mode, I use the software named "labelme" to label my data, I want to train a new model starting from pre-trained COCO weights, but I don't know how to write the annotation file in validate set, are there tools to help me? Thank you.

Validation annotation is the same as train

[gethubwy]

See information about pycococreator.

Hello, Excuse me, I want to know how to train a new mode, I use the software named "labelme" to label my data, I want to train a new model starting from pre-trained COCO weights, but I don't know how to write the annotation file in validate set, are there tools to help me? Thank you.

Validation annotation is the same as train

actually，they are different! for example:
instances_train2017.json:
{ "license": 1,
"file_name": "000000536292.jpg",
"coco_url": "http://images.cocodataset.org/train2017/000000536292.jpg
"height": 425,
"width": 640,
"date_captured": "2013-11-18 13:14:31
flickr_url": "http://farm4.staticflickr.com/3689/9154491588_866ac8791e_z.jpg",
"id": 536292
}

well the instances_val2017.json is:
{"segmentation": [
[314.85,365.65,305.09,379.74,295.88,378.12,293.17,393.29,357.11,390.58,391.24,387.87,382.03,365.65],
[289.38,375.95,266.63,370.53,262.29,388.95,271.5,393.29,280.17,392.2,282.34,392.2],
[317.01,341.27,308.89,330.44,306.18,320.68,307.8,312.56,311.6,288.72,332.73,283.3,350.06,289.26,350.06,300.1,352.77,305.51,351.15,327.73,349.52,339.11,351.15,342.9]],
"area": 4868.703549999998,
"iscrowd": 0,
"image_id": 545594,
"bbox": [262.29,283.3,128.95,109.99],
"category_id": 1,
"id": 207363
}

[konstantin-frolov]

In your example in validatiton annotations added only area field.
I use identical annotation for train and Val. It’s work fine.