Bounding boxes not showing

[aashay96]

Hey, I ran the code, and the bounding boxes show up for only very small thresholds like 0.007 that too randomly placed. Please help.

[subodh-malgonde]

Are your pre-processing the test images before passing them to the model?

You can have a look at Vehicle_Detection.ipynb, I have run the model on some test images and its working fine for me.

[aashay96]

Yes. I am doing pre- processing and using the same model as yours. The issue still exist. Can it have something to do with the weights?

[subodh-malgonde]

Could be. I hope you are using these weights https://drive.google.com/file/d/0B1tW_VtY7onibmdQWE1zVERxcjQ/view?usp=sharing

Are you using cv2 to read images or matplotlib? cv2 will read images in the BGR format whereas matplotlib will read them in RGB. The model is trained on RGB images. So you will have to ensure you are passing RGB image for testing.

[aashay96]

I am using matplotlib. I copied exactly your code, and used the mentioned weight, but the problem still exists.

[subodh-malgonde]

Can you upload your code somewhere and give me a link?

[aashay96]

sure. Here is the link to the code, though I copied your code, and tried, still the same issue.
http://paste.ubuntu.com/24253759/

[Angloher]

I am having the same issue

[ManjeeraJagiri]

Hey, I am facing the same issue. Did anyone figure out the problem?

[docPhil99]

I had this problem. I think the Keras 2 API is the problem. I used the solution here and now it works

[anandkoirala]

following worked for me
comment by mark-chan xslittlegrass/CarND-Vehicle-Detection#3
add this code before the load_weights(model,'./yolo-tiny.weights') function...

[ManjeeraJagiri]

@docPhil99 @anandkoirala Worked like a charm! Thanks a lot. Have any of you tried training with custom dataset?

[subodh-malgonde]

I would just like to mention that this repository uses keras 1.2.1 and tensorflow 0.12.1. If you use Keras 2.x or tensorflow 1.x you may need to modify the code.