search

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

Changing batch size #1124

Open Janzeero-PhD opened 4 years ago

Janzeero-PhD commented 4 years ago

Here is a big issue for whom are having not really good GPU. My certain yolov2.cfg does not contain information about batch size (only batch and subdivisions). Changing that, I again and again obtain a value 'batch size = 16' before training which fails. Convolutional algorithm fails or just OOM. Where to reduce batch size from 16 to 4 or 2? Cannot find file with this configuration. As well, I would be really appreciated for everybody who shares knowledge/link for better solutions for running such CNNs (Faster-RCNN, Yolo etc.) on my card. Google doesn't provide ready solution for searching this question, and StackOverFlow is tending to be silent.

HongYg commented 4 years ago

if you run at cmd , use this --batch 4 --batch 2

Janzeero-PhD commented 4 years ago

I ran from Jupyter Notebook. Ok, I have to review that.

Mohamed209 commented 4 years ago

@Janzeero-PhD You can still write cmds in notebook Just in new cell type ! your command