weights for models about darkflow HOT 13 CLOSED

Weights file are here http://pjreddie.com/darknet/yolo/
Look around for older version

from darkflow.

Do you have to keep the cfg and weights files paired? I would assume so. For example, I wouldn't use one of the darkflow cfg files and one of the pjreddie/darknet/yolo weight files, right?

from darkflow.

As long as --model, --load and --config is pointing to appropriate position, the pair .cfg and .weights are well-defined and there will be no confusion.

Two example that works fine:

./flow --model ~/darkflow/cfg/yolo.cfg --load ~/pjreddie/darknet/yolo.weights

# or partially load yolo-4c from yolo.weights, assuming yolo.weights' cfg is at ~/somewhereelse/
./flow --model ~/darkflow/cfg/yolo-4c.cfg --load ~/pjreddie/darknet/yolo.weights --config ~/somewhereelse/

from darkflow.

Good to know about the --config flag. I tried a test, using the darkflow tiny cfg, and the darknet V1 tiny cfg and weights.
Got an assertion error:

rob@skynet1:~/darkflow$ ./flow --model cfg/v1/yolo-tiny.cfg --train --load /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights --config /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.cfg --dataset /home/rob/Data_PascalVOC/VOCdevkit/VOC2012/JPEGImages/ --annotation /home/rob/Data_PascalVOC/VOCdevkit/VOC2012/Annotations/ --gpu 1.0
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcublas.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcudnn.so.5 locally
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcufft.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcurand.so.8.0 locally
Parsing cfg/v1/yolo-tiny.cfg
Loading /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights ...
Traceback (most recent call last):
File "./flow", line 42, in
tfnet = TFNet(FLAGS)
File "/home/rob/darkflow/net/build.py", line 34, in init
darknet = Darknet(FLAGS)
File "/home/rob/darkflow/dark/darknet.py", line 26, in init
self.load_weights()
File "/home/rob/darkflow/dark/darknet.py", line 78, in load_weights
wgts_loader = loader.create_loader(*args)
File "/home/rob/darkflow/utils/loader.py", line 104, in create_loader
return load_type(path, cfg)
File "/home/rob/darkflow/utils/loader.py", line 18, in init
self.load(*args)
File "/home/rob/darkflow/utils/loader.py", line 69, in load
val = walker.walk(new.wsize[par])
File "/home/rob/darkflow/utils/loader.py", line 126, in walk
'Over-read {}'.format(self.path)
AssertionError: Over-read /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights

Looking through what over-read means right now...

from darkflow.

BTW, pulling both --load and --config arguments out of the above call (starting from scratch weights basically) allows training to start fine.

from darkflow.

Your command:

./flow --model cfg/v1/yolo-tiny.cfg --load /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights --config /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.cfg

Is invalid, as --config should be a folder, to look for tiny-yolo.cfg, not the specific file, so the correct command is

./flow --model cfg/v1/yolo-tiny.cfg --load /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights --config /home/rob/models_darknet/yoloV1_tiny/

from darkflow.

Same error, using --config with folder instead of file. Error below.

rob@skynet1:~/darkflow$ ./flow --model cfg/v1/yolo-tiny.cfg --train --load /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights --config /home/rob/models_darknet/yoloV1_tiny/ --dataset /home/rob/Data_PascalVOC/VOCdevkit/VOC2012/JPEGImages/ --annotation /home/rob/Data_PascalVOC/VOCdevkit/VOC2012/Annotations/ --gpu 1.0
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcublas.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcudnn.so.5 locally
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcufft.so.8.0 locally
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:120] successfully opened CUDA library libcurand.so.8.0 locally
Parsing /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.cfg
Parsing cfg/v1/yolo-tiny.cfg
Loading /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights ...
Traceback (most recent call last):
File "./flow", line 42, in
tfnet = TFNet(FLAGS)
File "/home/rob/darkflow/net/build.py", line 34, in init
darknet = Darknet(FLAGS)
File "/home/rob/darkflow/dark/darknet.py", line 26, in init
self.load_weights()
File "/home/rob/darkflow/dark/darknet.py", line 78, in load_weights
wgts_loader = loader.create_loader(*args)
File "/home/rob/darkflow/utils/loader.py", line 104, in create_loader
return load_type(path, cfg)
File "/home/rob/darkflow/utils/loader.py", line 18, in init
self.load(*args)
File "/home/rob/darkflow/utils/loader.py", line 69, in load
val = walker.walk(new.wsize[par])
File "/home/rob/darkflow/utils/loader.py", line 126, in walk
'Over-read {}'.format(self.path)
AssertionError: Over-read /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights

from darkflow.

Seems like it is getting hung up on parsing the darknet weights file. Incidentally, would using the --binary flag work, instead of the --load flag?

from darkflow.

The message means home/rob/models_darknet/yoloV1_tiny/tiny-yolo.cfg indicates a larger size than /home/rob/models_darknet/yoloV1_tiny/tiny-yolo.weights actually is.

from darkflow.

Okay, interesting. Might be a problem with the files I pulled from the darknet site.
I suspected something similar, tried a different set (tiny_extract), with V1 extraction cfg and weights, and seems to work okay.

from darkflow.

I suppose you are trying to load yolo-tiny of YOLOv1.0 from tiny-yolo of YOLOv1.1 ? Doing that on my machine would be

./flow --model cfg/v1/yolo-tiny.cfg --load bin/v1.1/tiny-yolov1.weights --config cfg/v1.1/

And it works okay.

from darkflow.

Possibly. I'll take another look to make sure I'm using the correct pairing.

from darkflow.

hi,you,when I want to train a new model Iues the order:
./flow --model cfg/v3/tiny-yolo-voc-3c.cfg --load bin/tiny-yolo-voc.weights
#######tiny-yolo-voc is a copy of tiny-yolo-voc and class=3,filter=40,and label.txt include 3 class###
It come out a problem like this:
AssertionError: expect 63123056 bytes, found 63471556
I know it caused by the unpair between the .weight and .cfg ,how to solve this problem
thank you!

from darkflow.