Darkflow's tiny-yolo_v2.cfg is not equal to Darknet's tiny-yolo_v2.cfg , Why ?

thtrieu / darkflow

Translate darknet to tensorflow. Load trained weights, retrain/fine-tune using tensorflow, export constant graph def to mobile devices

GNU General Public License v3.0

6.13k stars 2.08k forks source link

Darkflow's tiny-yolo_v2.cfg is not equal to Darknet's tiny-yolo_v2.cfg , Why ? #857

Open LucasMahieu opened 6 years ago

LucasMahieu commented 6 years ago

I'm wondering why the darkflow's tiny yolov2 is not exactly the same as the original darknet one :

darknet tiny yolo v2
darkflow tiny yolo v2

As you can see, the anchors are not the same (by they are both trained one COCO). And the "filters" number of the before last convolution are not the same : 512 for darnet vs 1024 for darkflow.

Could anyone explain to me why this 2 cfg are different ?

Thanks !

LucasMahieu commented 5 years ago

It seens that there is also differences for Yolov2 version :

between here :

Darnet Yolo v2 cfg and
Darkflow Yolo v2 cfg

Just do a diff between these two files