andy-yun / pytorch-0.4-yolov3

Yet Another Implimentation of Pytroch 0.4.1 and YoloV3 on python3
MIT License
279 stars 72 forks source link

param about initial weights #21

Closed door5719 closed 6 years ago

door5719 commented 6 years ago

sir ,I want to know: python train.py -d cfg/coco.data -c cfg/yolo_v3.cfg -w yolov3.weights

-w means initial weights?

How can I train model without initial weights? Thank you sir~

andy-yun commented 6 years ago

sure. -w means the weights. In nowadays, the data increase. Therefore, many alogorithm uses initial weights for training. In yolo model. darknet19_448.conv.23 and darknet53.conv.74 are used for initial weights of yolov2 and yolov3, respectively.

These files are the pretrained weights for imagenet. you can download as follows. wget https://pjreddie.com/media/files/darknet19_448.conv.23 wget https://pjreddie.com/media/files/darknet53.conv.74

door5719 commented 6 years ago

Thank you sir for your reply. I want to train model without pre-trained weights ,random initial weights instead.

door5719 commented 6 years ago

I want to get a effective weights file of yolov3(coco_dateset80class), about how many epoch times will generate the weights file at least.

andy-yun commented 6 years ago

@door5719 I don't know. It may be necessary an analysis of training errors and validation performance. See https://github.com/AlexeyAB/darknet

door5719 commented 6 years ago

Thank you sir.