llSourcell / YOLO_Object_Detection

This is the code for "YOLO Object Detection" by Siraj Raval on Youtube
GNU General Public License v3.0
1.73k stars 795 forks source link

while training - loss parameter almost doesnt decrease #14

Open tamirm2009 opened 6 years ago

tamirm2009 commented 6 years ago

@llSourcell

Hi I'm running training process but loss parameter doesnt decreasing: first step it was 107% loss. On Step 425 loss is 102% only. What Should I do ? Is something suspicious/wrong ? I have 6 classes 1,2,3,4,5,6 . And dataset of 59 images only(XML file for each image). I'm training using yolov2-tiny-voc file --> edited to yolov2-tiny-voc-6c . Weights -> yolov2-tiny.weights(43896KB). BatchSize/Subdivision is relevant? [net] /# Testing batch=16 subdivisions=4 /# Training /# batch=64 /# subdivisions=2

I added / after the # because its bolding in github the text afterwards.(So my cfg file is without /).

cmdline run string: python flow --model cfg/yolov2-tiny-6c.cfg --load bin/yolov2-tiny.weights --train --annotation new_model_data/annotations --dataset new_model_data/images