search

mystic123 / tensorflow-yolo-v3

Implementation of YOLO v3 object detector in Tensorflow (TF-Slim)
https://medium.com/@pawekapica_31302/implementing-yolo-v3-in-tensorflow-tf-slim-c3c55ff59dbe
Apache License 2.0
890 stars 353 forks source link

Why Is Yolo V3 tiny's performance so bad? #47

Open gjraza opened 5 years ago

gjraza commented 5 years ago

I am loading Yolov3 tiny model and yolov3-tiny.weights downloaded from pjreddie's site in demo.py but I am not getting very good results. Is it actually that bad or am I doing something wrong? Have a look at the attached image.

fastt

gjraza commented 5 years ago

and its not even detecting this image. P.s I am using coco.names with it

plane

mystic123 commented 5 years ago

Can you run this on the pjreddie's implementation and upload results?

gjraza commented 5 years ago

Hi, I ran the original pjreddie's implementation and got the these results along with some other predictions. Looks like your implementation is under performing(?) As you can see, TF version is detecting only one person and the original version is detecting all persons plus a car. Its still not as accurate but the TF version is under performing I guess

predictions predictions1 predictions2 predictions3

gjraza commented 5 years ago

it still couldnt detect all planes though. So I guess this behavior is predictable considering its mAP of 33.1 on COCO