How to convert yolo layer output to bounding box coordinates?

[DJ Lee]

yolo layers seem to have an output shape of

[ batch, height, width, 3  * (5 + num_classes)]

where height and width are size of the grid.

this is what I've tried referencing model.py

def get_bbox_with_anchors(prediction, anchors, classes_num=8):
    # split by anchor box
    # split into three
    split_preds = tf.split(prediction, 3, axis=-1)
    pred = tf.stack(split_preds, axis=-2)
    
    # get grid shape
    grid_size = tf.shape(pred)[1]
    box_xy, box_wh, objectness, class_probs = tf.split(
        pred, (2, 2, 1, classes_num), axis=-1)
    
    # apply transforms
    # box_xy = tf.sigmoid(box_xy)
    # darknet does not seem to appy sigmoid on offset
    objectness = tf.sigmoid(objectness)
    class_probs = tf.sigmoid(class_probs)
    pred_box = tf.concat((box_xy, box_wh), axis=-1)  # original xywh for loss

    # calculate offsets
    grid = tf.meshgrid(tf.range(grid_size), tf.range(grid_size))
    grid = tf.expand_dims(tf.stack(grid, axis=-1), axis=2)  # [gx, gy, 1, 2]

    box_xy = (box_xy + tf.cast(grid, tf.float32)) / \
        tf.cast(grid_size, tf.float32)
    box_wh = tf.exp(box_wh) * anchors

    box_x1y1 = box_xy - box_wh / 2
    box_x2y2 = box_xy + box_wh / 2
    bbox = tf.concat([box_x1y1, box_x2y2], axis=-1)

    return bbox, objectness, class_probs

overlaying these bbox to the original image seems to be off quite a bit
I am not sure what I am doing wrong.
Any help is appreciated

[nishanthcgit]

@greyowl Hey, I'm sort of new to TF, did you resolve this finally? How did you get the actual bbox coordinates from the output layers?

[DJ Lee]

I don't have the code but I ended up implementing the output -> bbox coordinates in tensorflow

[nishanthcgit]

I'll look that up, thanks a lot!

[Robin2091]

@nishanthcgit Hi, were you able to figure out how to get bounding boxes? I tried referencing other repositories including model.py, but I always got zero bounding boxes. Any help will be appreciated. Thanks

[nishanthcgit]

I had looked into how tensorflow itself does the output -> coordinates transformation in its own models - its not actually too hard, just poke around in all the functions called. I don't have the exact name of the function now but I had found one in there that does the same and used it..
Check out the tensorflow object detection API, it is 100% in there, I am just not sure of the exact function name now