search

arduino-libraries / Arduino_OV767X

Official OV767X Library for Arduino , currently supports OV7670 and OV7675 cameras
https://www.arduino.cc/reference/en/libraries/arduino_ov767x/
GNU General Public License v2.0
68 stars 25 forks source link

Feature Request. Any chance of QQQVGA Resolution? #8

Closed hpssjellis closed 3 years ago

hpssjellis commented 3 years ago

Any chance of getting a few more resolutions out of the OV7670 Camera module. Here is a list of what OpenMV has going, For a bit of machine learning I would just like a few sizes smaller than what we already have, such as:

QQQVGA: 80x60 QQQQVGA: 40x30

Note: Smaller is faster and easier to process.

Data from OpenMV

sensor.QQCIF: 88x72 sensor.QCIF: 176x144 sensor.CIF: 352x288 sensor.QQSIF: 88x60 sensor.QSIF: 176x120 sensor.SIF: 352x240 sensor.QQQQVGA: 40x30 sensor.QQQVGA: 80x60 sensor.QQVGA: 160x120 sensor.QVGA: 320x240 sensor.VGA: 640x480 sensor.HQQQVGA: 80x40 sensor.HQQVGA: 160x80 sensor.HQVGA: 240x160 sensor.B64X32: 64x32 (for use with image.find_displacement()) sensor.B64X64: 64x64 (for use with image.find_displacement()) sensor.B128X64: 128x64 (for use with image.find_displacement()) sensor.B128X128: 128x128 (for use with image.find_displacement()) sensor.LCD: 128x160 (for use with the lcd shield) sensor.QQVGA2: 128x160 (for use with the lcd shield) sensor.WVGA: 720x480 (for the MT9V034) sensor.WVGA2:752x480 (for the MT9V034) sensor.SVGA: 800x600 (only for the OV2640/OV5640 sensor) sensor.XGA: 1024x768 (only for the OV2640/OV5640 sensor) sensor.SXGA: 1280x1024 (only for the OV2640/OV5640 sensor) sensor.UXGA: 1600x1200 (only for the OV2640/OV5640 sensor) sensor.HD: 1280x720 (only for the OV2640/OV5640 sensor) sensor.FHD: 1920x1080 (only for the OV5640 sensor) sensor.QHD: 2560x1440 (only for the OV5640 sensor) sensor.QXGA: 2048x1536 (only for the OV5640 sensor) sensor.WQXGA: 2560x1600 (only for the OV5640 sensor) sensor.WQXGA2: 2592x1944 (only for the OV5640 sensor)

lightb0x commented 3 years ago

64x64 and 96x96 support would be lovely for ML applications. Transfer learning from imagenet_resized, pretrained MobileNet V2 will be possible.

hpssjellis commented 3 years ago

Looking at the code it doesn't seem like it would be that hard to do QQQVGA or QQQQVGA Code sample here

This looks like the main entry point for QQVGA 160 x 120 QQQVGA 80 x 60 QQQQVGA 40 x 30

    case QQVGA:
      _width = 160;
      _height = 120;
      break;
facchinm commented 3 years ago

@hpssjellis that is the easy part :slightly_smiling_face: Then the sensor must be configured with the right resolution using ov7670_configure (since the nRF52 has no hardware crop, Portenta has and uses it). Formars are then converted using these tables (https://github.com/arduino-libraries/Arduino_OV767X/blob/30668622f7e98c4fd248ee005a6bdcee37948b10/src/utility/ov7670.c#L835-L892); most of the values are empirically determined since Ominivision datasheets are simply filled with errors and incorrect information. So, it's a big effort. Anyway, if anyone is willing to take it I'll be glad to support.

mbanzi commented 3 years ago

BTW there is a new pull request which adds a number of resolutions to the library, check it out here