search

philipperemy / yolo-9000

YOLO9000: Better, Faster, Stronger - Real-Time Object Detection. 9000 classes!
Apache License 2.0
1.18k stars 309 forks source link

About the names #15

Open hengshan123 opened 6 years ago

hengshan123 commented 6 years ago

I run the "./darknet detector test cfg/combine9k.data cfg/yolo9000.cfg ../yolo9000-weights/yolo9000.weights data/person.jpg" command. I can get the detection,but the names is not the same with the README. for example, i get worker instead of 'goat herder' , horse instead of shetland pony hunting dog instead of Gorden setter. Is this project updated or something is not right ?

stesteau commented 6 years ago

I can't figure out why but names are correct when you compile the CPU version...

philipperemy commented 6 years ago

Hum yes I found out that the results can be slightly different when running on a CPU and GPU. If you can figure out why, I'd be def interested to know! Maybe a different seed, rounding values...

xiaoxiongli commented 5 years ago

@hengshanji @moskiteau @philipperemy i met the same issue that when compile with GPU the name it not correct, and when compile with CPU, the result is right. I wonder How do you fix it?

philipperemy commented 5 years ago

Hum not sure why. This repo does not contain the code of darknet. It's more a demo to show YOLO with 9000 classes. Refer to: https://pjreddie.com/darknet/yolo/ @xiaoxiongli

huangeason26 commented 4 years ago

@hengshanji @moskiteau @philipperemy @xiaoxiongli i met the same issue, i wonder if any of you had figure it out?

sohambannerjee8 commented 4 years ago

Same Issue.Even I found some images are giving empty result. I have created one issue also regarding this