search

qqwweee / keras-yolo3

A Keras implementation of YOLOv3 (Tensorflow backend)
MIT License
7.14k stars 3.44k forks source link

why the speed of videos only 1 fps #72

Closed ll1214 closed 6 years ago

ll1214 commented 6 years ago

it so slow!

qqwweee commented 6 years ago

Use your GPU.

ll1214 commented 6 years ago

thank you very much

ll1214 commented 6 years ago

But how to set GPU to run it ,because my environment is tensorflow-gpu and it can noly use GPU

ll1214 commented 6 years ago

And I haved just uninstall the Theano，but it didn't work

qqwweee commented 6 years ago

Did you install CUDA and cuDNN?

ll1214 commented 6 years ago

Yes I have installed

ll1214 commented 6 years ago

And My computer graphics card is GTX960M

qqwweee commented 6 years ago

The device information is displayed on the screen when you run tensorflow. Read it and find what happened.

ll1214 commented 6 years ago

I shows thUsing TensorFlow backend. model_data/yolo.h5 model, anchors, and classes loaded. (416, 416, 3) Found 9 boxes for img truck 0.36 (361, 122) (458, 203) truck 0.48 (368, 95) (498, 186) car 0.32 (201, 125) (356, 191) car 0.51 (244, 128) (406, 205) car 0.53 (361, 122) (458, 203) car 0.96 (694, 152) (745, 189) car 0.98 (624, 153) (697, 204) car 0.99 (183, 149) (336, 287) person 0.33 (529, 145) (543, 175) 2.37719493887547 (416, 416, 3) Found 6 boxes for img truck 0.57 (335, 90) (479, 181) car 0.86 (309, 127) (420, 208) car 0.88 (197, 117) (317, 222) car 0.95 (623, 152) (696, 201) car 0.96 (703, 150) (755, 187) person 0.43 (533, 147) (548, 173) 1.6571010365649785 (416, 416, 3)at :

ll1214 commented 6 years ago

Another idea is whether I can train a model to identify traffic signs and objects at the same time.

ll1214 commented 6 years ago

The traffic sign data set is in http://cg.cs.tsinghua.edu.cn/traffic-sign/, which is a data set rented by Tsinghua University.

ll1214 commented 6 years ago

The useful information I got from above is that the prediction time for each frame is 1.7s.

qqwweee commented 6 years ago

Did GPU work before? Perhaps the environment is not correctly configured. Search the internet for solution.

Traffic sign is more like Vehicle License Plate Recognition and is unsuited to Object Detection, I think.

ll1214 commented 6 years ago

Yes I thank you are right ,I just uninstalled tensorflow-gpu and install tensorflow.The speed is same.And I run it in windows

ll1214 commented 6 years ago

But my idea is to make yolov3 identify traffic signs on general testing, but it only adds a few more categories.

qqwweee commented 6 years ago

Then it is feasible someway. Just make a try.

TommyAnqi commented 6 years ago

Most of the time consumed on image preprocessing as python script is slow. Only print the time on the forward part

ll1214 commented 6 years ago

Thanks for your help,and maybe we could make friends.I have recently studied the analysis and prediction of medical imaging. Lung cancer prediction in kaggle games