Darkflow: ValueError: cannot convert float NaN to integer

[jiankang1991]

Dear
Thank you very much for posting Yolo in tensorflow.
I try the demo and get the following error. Could you please have a look?
First, I download the yolotiny.weights from Yolo website.

Then,

python clean.py /home/karl/Documents/VOCdevkit/VOC2012/Annotations
[===================>]100%
Statistics:
pottedplant: 13442
person: 17401
tvmonitor: 15512
Dataset size: 26089

At last, I run the test code:

python tensor.py --test data --model tiny
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcublas.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcudnn.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcufft.so locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcuda.so.1 locally
I tensorflow/stream_executor/dso_loader.cc:108] successfully opened CUDA library libcurand.so locally
parsing yolo-tiny.cfg
Traceback (most recent call last):
  File "tensor.py", line 37, in <module>
    yoloNet = YOLO(FLAGS.model + int(step > 0) * '-{}'.format(step))
  File "/home/karl/Documents/online_code/yolotf/Yolo.py", line 57, in __init__
    self.build(model)
  File "/home/karl/Documents/online_code/yolotf/Yolo.py", line 71, in build
    for i, info in enumerate(layers):
  File "/home/karl/Documents/online_code/yolotf/configs/process.py", line 60, in cfg_yielder
    size = int(size)
ValueError: cannot convert float NaN to integer

How to fix that?
Thank you very much.

[jiankang1991]

The size of yolotiny.weights is 103.4Mb and I think the error comes out of this. But I do not know how to fix that.

[thtrieu]

Hi @karlTUM, according to the available configs/yolo-tiny.cfg, the size of yolo-tiny.weights MUST be 180MB (4 bytes for each parameter - you can check it yourself). So the problem is either you downloaded the wrong weight file, or the YOLO authors released a new config with size 103MB. In the latter situation, you must replace this new config into ./configs before running the demo.

Good luck!

[thtrieu]

I've looked at the YOLO webpage. They indeed released a new config compare to when I develop this repo. So you must use the old weight file (180MB) for the demo to be successful.

[jiankang1991]

Thank you. I will download an old version

[kivijoshi]

where can I download old weight file?

[thtrieu]

@kivijoshi I'll try to upload it.

[kivijoshi]

thanks :)

[kivijoshi]

I also tried swapping cfg file with new one from darknet repo. but it does not work
I get this msg..
expect 870256 bytes, found 108399816

[thtrieu]

new config file consists of batch-normalization layers. This is new compare to the old configs and my code currently cannot work with these.

[thtrieu]

@karlTUM @kivijoshi here is the old weight file https://drive.google.com/open?id=0B1tW_VtY7onibmdQWE1zVERxcjQ

[jwnsu]

Thanks a lot for the thread. I'm wondering whether anyone has the old yolo-full weight file?

The old tiny weight file works perfectly fine.

[thtrieu]

@jwnsu I have that one too, please wait until I uploaded it.

[jwnsu]

@thtrieu I'm wondering whether you find time to upload the old full yolo weight files. From the testing, it seems full model outperforms small and tiny quite a bit in a few test models.

When I create new model (e.g. 3c), tried to copy extraction.conv.weights as yolo-full.weights, but it seems genw.py does not take partial weight as darknet does.

Thanks again. Appreciate the project and efforts.

[Mu7ammad]

@thtrieu I'd like also to ask about the old yolo-coco.weights file since the current one online has been replaced with batch-normalization conv layers too. This will complete the configuration set for this fine project. Thanks in advance.

[thtrieu]

@jwnsu @Mu7ammad
The full weight file is here: https://drive.google.com/open?id=0B1tW_VtY7onicFlqclhnRGlodGM

I believe I will finish building the batch-normalization layer as well as allowing to import partial network within this week. Thanks for pointing out the need.

P/S: the old COCO model is one that I did not have access to unfortunately :(

[jwnsu]

Thanks a lot, Trieu. One question for you: do you have plan to refresh
yolotf to pick up recent changes in yolo, e.g. the batch_normalize feature.

Best,
Dennis

On Mon, Nov 7, 2016 at 9:10 AM, Trieu [email protected] wrote:

@jwnsu https://github.com/jwnsu @Mu7ammad https://github.com/Mu7ammad
The full weight file is here: https://drive.google.com/open?id=0B1tW_
VtY7onicFlqclhnRGlodGM

I believe I will finish building the batch-normalization layer as well as
allowing to import partial network within this week. Thanks for pointing
out the need.

P/S: the old COCO model is one that I did not have access to unfortunately
:(

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/thtrieu/yolotf/issues/1#issuecomment-258897563, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAM4Ky1cHvisw2gd9Uc3AHIULuLleQruks5q71uSgaJpZM4KFoaR
.

[thtrieu]

@jwnsu I do, the plan is to support partial weight file as well as adding batch_norm layer within this week, as stated in the previous post

[jwnsu]

Great! Will watch your checkin, I'm happy to test and report result to you.

Dennis

On Mon, Nov 7, 2016 at 9:48 AM, Trieu [email protected] wrote:

@jwnsu https://github.com/jwnsu I do, the plan is to support partial
weight file as well as adding batch_norm layer within this week, as stated
in the previous post

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
https://github.com/thtrieu/yolotf/issues/1#issuecomment-258908020, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AAM4K9Dpgw87zobCGHgl1434v6TvCevjks5q72RlgaJpZM4KFoaR
.

[Mu7ammad]

@thtrieu That would be great, looking forward to these updates, 頑張って！

[thtrieu]

@jwnsu @Mu7ammad
Sorry that I missed my deadline, but I finish it just few minutes ago and want to give you guys a friendly ping. Now everything is fine with the new tiny-yolo (108MB) and yolo (789 MB). The old configs are still working fine (yolo-full 1GB, yolo-small 376MB and yolo-tiny 180MB). A lot of code is refactored and so does the syntax. So have a quick look at README to be updated.

The next goal is to work on yolov2, or essentially work on route, reorg and region layers.

Cheers.

[Mu7ammad]

@thtrieu great news it now supports the latest models, thanks for the good effort. I'm then interested to try out tiny-coco next since it's close to tiny-yolo, will explore the code and syntax.

[thtrieu]

@Mu7ammad The dynamics are already there. The only thing you need to do is to add coco label list, which I've already done in a new commit.

Now you can run coco (tiny-coco and yolo-coco) just fine. (tell me if there's any problem)