Yolo v3 training on coco data set

Hi,

I want to train the the yolo v3 model with coco dataset. how can i do that?

Thanks

Answer 1

[T.Nikhil kumar]

Hey Asif !

I understand that you want to build a yolov3 object detector model and train it on the COCO dataset.

There are pretrained YOLOv3 object detectors trained on COCO dataset. You do not need to train a network separately. The following command lets you create a detector using YOLO v3 deep learning networks trained on a COCO dataset.

detector = yolov3ObjectDetector(name)

Here, name is the name of the pretrained YOLO v3 deep learning network, specified as one of these:

• 'darknet53-coco' — A pretrained YOLO v3 deep learning network created using DarkNet-53 as the base network and trained on COCO dataset.
• 'tiny-yolov3-coco' — A pretrained YOLO v3 deep learning network created using a small base network and trained on COCO dataset.

For reference ,please go through

(Link:https://www.mathworks.com/help/vision/ref/yolov3objectdetector.html)

If you still wish to perform training on your own then please refer the following example

(Link:https://www.mathworks.com/help/vision/ug/object-detection-using-yolo-v3-deep-learning.html#ObjectDetectionUsingYOLOV3DeepLearningExample-9)

Answer 2

[Divya Gaddipati]

You can refer to the following link for training a YOLOv3 object detector. In place of the dataset used in this example, you can load your own dataset and arrange it in the same format as described in the example

https://www.mathworks.com/help/vision/examples/object-detection-using-yolo-v3-deep-learning.html

[cui]

This official example cellfun function is not recommended, and it is better to support custom building yolov3Layer.

Answer 3

[cui]

This is the yolov3 you want, but there is a problem with saving the model during training, especially the parameter saving of the bn layer should be consistent with darknet, and the labeled [x, y, w, h], instead of Normalized [center_x, center_y, w, h ]

cocoAPI, https://github.com/cuixing158/cocoAPI