Recently doing research project, need to use target identification, heard from a friend that yolo, found after the search is in line with our requirements, today try to use a bit on the computer, I feel pretty good, mainly the time spent on configuration.
Most of them are pressing to do this blog, blogger wrote in great detail:
https://blog.csdn.net/weixin_44684139/article/details/101712290
to re-file the relatively large right note, if downloaded from the official website might be very for a long time, you can search for download on CSDN.
The main course record at the configuration of (spent the afternoon):
a very important point: tensorflow with ten million 2 1 Do not use!
tensorflow with ten million 2 1 Do not use!
tensorflow with ten million 2 1 Do not use!
The important thing to say three times, tf2 just came out last year, session what changed substantially, yolo on github code is not supported.
My tf-gpu would have been 2.0, but after downloading yolo not be the last step, plus once too often wrong, huff, unloaded downgrade!
After unloading I installed version 1.12.0 of tf-gpu, then surprised to find my cuda wrong version, you have to re whole. . .
Finally, I configured to: tf-gpu 1.12.0 cuda: 9.0 cuDNN: 7
when the next cuda is a tip, the official website to download from slow, you can right copy the link, download Thunder
Thunder on these few comments on the very taste (manual nugget)
finally finished the whole cuda found keras version not right, had to unload the heavy equipment, so children's shoes should pay attention
tf (gpu) -cuda-cuDNN- keras the version corresponding relationship! !
Finally, the working people under the tireless efforts and hard work, yolo operating environment and finally a good distribution! !
Can not wait to try a bit
input and output images as follows (with pictures and load weights may have to wait for some time, do not hurry):
final results also can not be identified and some may be squeezed together or picture clarity is not high.
yolov3 is very interesting, recent outbreaks so much, my friends at home idle is idle, you can try to run it themselves.
