AlexeyAB
commented
6 years ago For training yolo-voc.2.0.cfg you should use darknet19_448.conv.23.
But for training tiny-yolo-voc.cfg you can train without pre-trained weights or you should use tiny-yolo-voc.conv.13: https://github.com/AlexeyAB/darknet/issues/389#issuecomment-365925496
I successfully ran with cuDNN: darknet.exe classifier predict cfg/imagenet1k.data cfg/darknet19_448.cfg darknet19_448.weights
Using Windows 7x64, CUDA 9.1 and cuDNN 7. What CUDA and cuDNN version do you use, and what GPU do you use?
D:\Darknet2\darknet\build\darknet\x64>darknet.exe classifier predict cfg/imagene
t1k.data cfg/darknet19_448.cfg darknet19_448.weights
layer filters size input output
0 conv 32 3 x 3 / 1 448 x 448 x 3 -> 448 x 448 x 32
1 max 2 x 2 / 2 448 x 448 x 32 -> 224 x 224 x 32
2 conv 64 3 x 3 / 1 224 x 224 x 32 -> 224 x 224 x 64
3 max 2 x 2 / 2 224 x 224 x 64 -> 112 x 112 x 64
4 conv 128 3 x 3 / 1 112 x 112 x 64 -> 112 x 112 x 128
5 conv 64 1 x 1 / 1 112 x 112 x 128 -> 112 x 112 x 64
6 conv 128 3 x 3 / 1 112 x 112 x 64 -> 112 x 112 x 128
7 max 2 x 2 / 2 112 x 112 x 128 -> 56 x 56 x 128
8 conv 256 3 x 3 / 1 56 x 56 x 128 -> 56 x 56 x 256
9 conv 128 1 x 1 / 1 56 x 56 x 256 -> 56 x 56 x 128
10 conv 256 3 x 3 / 1 56 x 56 x 128 -> 56 x 56 x 256
11 max 2 x 2 / 2 56 x 56 x 256 -> 28 x 28 x 256
12 conv 512 3 x 3 / 1 28 x 28 x 256 -> 28 x 28 x 512
13 conv 256 1 x 1 / 1 28 x 28 x 512 -> 28 x 28 x 256
14 conv 512 3 x 3 / 1 28 x 28 x 256 -> 28 x 28 x 512
15 conv 256 1 x 1 / 1 28 x 28 x 512 -> 28 x 28 x 256
16 conv 512 3 x 3 / 1 28 x 28 x 256 -> 28 x 28 x 512
17 max 2 x 2 / 2 28 x 28 x 512 -> 14 x 14 x 512
18 conv 1024 3 x 3 / 1 14 x 14 x 512 -> 14 x 14 x1024
19 conv 512 1 x 1 / 1 14 x 14 x1024 -> 14 x 14 x 512
20 conv 1024 3 x 3 / 1 14 x 14 x 512 -> 14 x 14 x1024
21 conv 512 1 x 1 / 1 14 x 14 x1024 -> 14 x 14 x 512
22 conv 1024 3 x 3 / 1 14 x 14 x 512 -> 14 x 14 x1024
23 conv 1000 1 x 1 / 1 14 x 14 x1024 -> 14 x 14 x1000
24 avg 14 x 14 x1000 -> 1000
25 softmax 1000
26 cost 1000
Loading weights from darknet19_448.weights...
seen 32
Done!
Enter Image Path: dog.jpg
448 448
dog.jpg: Predicted in 0.020000 seconds.
malamute: 0.841539
Eskimo dog: 0.044991
Siberian husky: 0.026070
dogsled: 0.006565
keeshond: 0.004889
Enter Image Path:
When I test classifier with cudnn=1, extraction.weights, darknet19.weights etc are giving me wrong classification but as I change my makefile to cudnn=0, they give me right classification. So my question is that will it worsen the pre-trained model if I train my detector even with darknet19_448.conv.23 weights with cudnn=1 . Regards