mismatch between darket output and tensorflow-yolo-v3 output

[hellbago]

I tried to compare the output of yolo-darknet and your version of yolo implemented in tensorflow in the following image. As you can see there is a clear mismatch between the two. I'n using tha same confidence and non-max suppresion thresholds. Further I'm using tensorflow 1.4 with cuda 8.0 on my laptop. Any comments about this? @mystic123 If you want to try by yourself this is the link of image that i've used: https://media.wired.com/photos/5afb61bd57c02d59cb85a80e/master/w_628,c_limit/GettyImages-469981581.jpg

Output from tensorflow-yolo-v3: Output from darknet:

[alexanderfrey]

@hellbago Did you use the latest version ?

[hellbago]

Do you mean the latest version of the code of this repo or the latest version of tensorflow? @alexanderfrey

[sagarkar10]

Also having similar issue, can someone help me to compare the output of each layer... In my tesing, after the route layer is applied the outputs are not matching....

[YunYang1994]

I think it's related to NMS algorithmn. hope it helps you https://github.com/YunYang1994/tensorflow-yolov3

[LucasMahieu]

I also have this error ! @mystic123 any idea ? I think it is related to this : https://github.com/mystic123/tensorflow-yolo-v3/issues/50 and https://github.com/mystic123/tensorflow-yolo-v3/issues/51

[mystic123]

Sorry, I have no idea :/ I thought I this code works exactly the same as original implementation, but I may be wrong. All PRs with fixes are appreciated :)

[mystic123]

Can someone check if this works with the latest PR merged (https://github.com/mystic123/tensorflow-yolo-v3/pull/49)?