real-time object recognization using digits model (web camera)

[Gloriabhsfer]

Hi all,

I want to train a deep learning model and do object classification and detection for my own object. I trained my dataset through digits-server and get my Caffe model.

Because I want to use this model to do real-time object detection and classification, to get my object information and position, I need to use a web camera as my input to the training model. In your example, there only have a program for the image. I have tested it and it works fine. So I want to modify the code and replace the image tag with the cv2.capture frame in OpenCV.

It seems my format for my video is not the format in your Caffe model. I use google net to train my model, and my web camera is logi 720p, which output format is YUVY 2,2,4. I know the format in google net is RGB and need to transfer the OpenCV mat format into np.darray. And I have tested them but didn't work.

Can you give me some ideas to modify it? Or because my training data is the image, so I cannot apply a video on the model. If so, I will test other methods to do object recognition.

Thanks, Gloria