search

Ma-Dan / keras-yolo4

A Keras implementation of YOLOv4 (Tensorflow backend)
MIT License
382 stars 175 forks source link

GPU memory requirement #11

Open timezone163 opened 4 years ago

timezone163 commented 4 years ago

请问训练该模型使用的GPU是什么型号，显存大小？ GPU=8G显存，能否支持训练batch_size = 4？

Simon-liusheng commented 4 years ago

支持不了的

timezone163 commented 4 years ago

谢谢您的回答，请问作者使用的是什么显卡，准备入手一款训练用显卡，显存最低要求，谢谢

Simon-liusheng commented 4 years ago

我现在的就是8g，但是bs只能2，我还想用下百度的显卡，16g，配环境相当的麻烦。

timezone163 commented 4 years ago

@Simon-liusheng 8G显存的话，只需要将batch_size = 2就能完成作者的训练吗？model size不要修改吧？比如将（608，608）改为（512， 512），或者（416，416）。 因为我看默认下载的yolov4.weights的model size是（608，608）。 谢谢

robisen1 commented 4 years ago

@Simon-liusheng 8G显存的话，只需要将batch_size = 2就能完成作者的训练吗？model size不要修改吧？比如将（608，608）改为（512， 512），或者（416，416）。 因为我看默认下载的yolov4.weights的model size是（608，608）。 谢谢

ohh i did not notice that. are we supposed to be using 608X608 in train? I sort of assumed it would not matter if this was from Darknet yolov4. Also I have a RTX 2080 TI founders edition with 11g. It can barely handle a batch size of 4 on a small data set. This seems odd. Why is memory usage so much?

iliask97 commented 4 years ago

@Simon-liusheng 8G显存的话，只需要将batch_size = 2就能完成作者的训练吗？model size不要修改吧？比如将（608，608）改为（512， 512），或者（416，416）。 因为我看默认下载的yolov4.weights的model size是（608，608）。 谢谢

ohh i did not notice that. are we supposed to be using 608X608 in train? I sort of assumed it would not matter if this was from Darknet yolov4. Also I have a RTX 2080 TI founders edition with 11g. It can barely handle a batch size of 4 on a small data set. This seems odd. Why is memory usage so much?

qqwweee's implemantation is on yolov3 not v4. yolov4 is a way bigger network and requiers a lot more vram to run. In the github repository of the original yolo their is a section witch tells you what changes you need to do according to your gpu memory to train it faster