Yolo on VOC data

[tiagojdias]

Hi,

So i follow the stages stated on the website to train the VOC data with yolo. Even with GPU (Geforce GTX 1070) the training procedure took me about one day. So, i saw a lot of discussion on the about what config to use ... First i tried with the yolo-voc.cfg and then with yolo-voc.2.0.cfg ... although i'm not obtaining any labels for the dog figure (which was supposed since the network contains that class)

For training i've just used the normal command: ./darknet detector train data/voc.data yolo-voc.2.0.cfg darknet19_448.conv.23 and, then, ./darknet detector test cfg/voc.data cfg/yolo-voc.2.0.cfg backup/yolo-voc_final.weights data/dog.jpg not getting any labels.

Can you pls explain me what i'm doing wrong ...

[AlexeyAB]

HI,

• What values of batch and subdivision parameters did you use in your yolo-voc.2.0.cfg file?
• Do you use cuDNN in your Makefile?
• What OpenCV version do you use?
• What repository do you use, this fork or original repo https://github.com/pjreddie/darknet ?

[tiagojdias]

• What values of batch and subdivision parameters did you use in your yolo-voc.2.0.cfg file? A: batch = 64 ; subdivision = 8
• Do you use cuDNN in your Makefile? A: Yes, i enable it i'm using cuDNN 7.0 alongside with CUDA 8.0
• What OpenCV version do you use? A: 3.2.0
• What repository do you use, this fork or original repo https://github.com/pjreddie/darknet ? A: Yes, using the original one

[AlexeyAB]

• Try to detect using your trained model backup/yolo-voc_final.weights on this repo on Linux.
• Then if it doesn't work, try to compile this fork without cuDNN (CUDNN=0 in the Makefile)

[tiagojdias]

Hi,

Thank you for the sugestions but it didn't work. I'm supposing that for test i need to change the config file for batch=1 and subdivision=1... but i was not able to make it. I'll re-train from scratch and then i'll give you feedback.

2017-10-19 13:29 GMT+01:00 Alexey notifications@github.com:

• Try to detect using your trained model backup/yolo-voc_final.weights on this repo on Linux.
• Then if it doesn't work, try to compile this fork without cuDNN (CUDNN=0 in the Makefile)

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/AlexeyAB/darknet/issues/235#issuecomment-337892411, or mute the thread https://github.com/notifications/unsubscribe-auth/AMGFRv7_G1Dx2aAm-MepeJbS8wrzdrxJks5st0C4gaJpZM4P_A87 .

[AlexeyAB]

I'm supposing that for test i need to change the config file for batch=1 and subdivision=1

This is not necessary, this only reduces memory consumption for test.

But you must use batch=64 subdivision=8 for training, as described here: https://github.com/AlexeyAB/darknet#how-to-train-pascal-voc-data

[iraadit]

Should we use yolo-voc.cfg or yolo-voc.2.0.cfg?

On the website, for YOLOv2, it still links to yolo-voc.cfg

Is yolo-voc.cfg config file corresponding to the YOLOv2 / YOLO9000 paper? And yolo-voc.2.0.cfg to an updated (and better?) config?

[AlexeyAB]

@iraadit yolo-voc.cfg slightly better than yolo-voc.2.0.cfg, but some people have problems with training yolo-voc.cfg.

1. Yolo v2: yolo.2.0.cfg and yolo-voc.2.0.cfg - (since 17 Nov 2016): https://arxiv.org/pdf/1612.08242.pdf
2. Yolo v2.x: yolo.cfg and yolo-voc.cfg - (since 10 Apr 2017): https://pjreddie.com/darknet/yolo/

[iraadit]

Thank you !