apivovarov opened a new pull request #8477:
URL: https://github.com/apache/tvm/pull/8477


   This PR fixes darknet yolo bbox calculation in case network heigh and width 
are not equal.
   I have custom darknet yolo model with shape `[1,1,224,384]`
   Without this fix the boxes calculation gets outside of the image boundaries.
   tvm_output contains layer_out["output"]. Its shape format is `[_,_,h,w]` 
(variable `data` in this PR)
   But `_get_box` method signature uses  `lw, lh` parameters sequence. 
`_get_box(data, biases, n, location, lw, lh, w, h)`.
   the fix maps `lw` to `data[3]` and `lh` to `data[2]`.
   The problem was not detected earlier because vanilla yolov2 and yolov3 
models network usually have the same Height and Width. For example 
`[1,3,416,416]` or `[1,3,608,608]`


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to