Welcome new user! You can search existing questions and answers without registering, but please register to post new questions and receive answers. Note that due to large amounts of spam attempts, your first three posts will be manually moderated, so please be patient.
Because of un-manageable amounts of spam despite our use of CAPTCHAs, email authorization, and other tools, we have discontinued this forum (see the 700k+ registered users with validated email addresses at right?). Please email us any questions or post bug reports and feature requests on GitHub at https://github.com/jevois -- The content below remains available for future reference.
Welcome to JeVois Tech Zone, where you can ask questions and receive answers from other members of the community.

it is possible to make tensor flow recognition in hither resolutions ?

0 votes
it is possible to make tensor flow recognition in hither resolutions for example 800x600 OV2640 sensor ?
asked Aug 1, 2019 in Hardware Questions by tarzanboy13 (200 points)

1 Answer

0 votes

I depends on what you are trying to achieve. The TensorFlow models usually are trained for a specific resolution, eg 224x224 or 128x128

In the TensorFlowEasy module http://jevois.org/moddoc/TensorFlowEasy/modinfo.html we first take a central crop of the input video (the grey square shown in the screenshots), then resize this, if needed, to the network's expected input size, then process through the network. The crop size is determined by parameter "foa" with this doc:

Width and height (in pixels) of the fixed, central focus of attention. This is the size of the central image crop that is taken in each frame and fed to the deep neural network. If the foa size does not fit within the camera input frame size, it will be shrunk to fit. To avoid spending CPU resources on rescaling the selected image region, it is best to use here the size that the deep network expects as input.

So if you run at 800x600, you could set foa to 500 500 or whatever else you like. This is best done in params.cfg (see above link).

If you have a trained network that expects 800x600 inputs, then set foa to 800 600 too. The only thing there is that likely your network will run very slowly on JeVois and you might run out of memory depending on how complex your network is.
answered Aug 1, 2019 by JeVois (46,580 points)
other networks was trained with 1.2 million training images
my dataset is only 7000 images, do you think, that CPU and RAM will good ?
it's more a matter of how many layers, how many filters in each layer, etc

If your weight file is on the order of a few MB but not on the order of 100MB+ you should be ok. Our mobilenets have tflite files on the order of 1MB to 25MB and run fine. Even inception V3 at 91M works but is a bit slow.