search

HimaxWiseEyePlus / Seeed_Grove_Vision_AI_Module_V2

https://www.himax.com.tw/products/wiseeye-ai-sensing/wiseeye2-ai-processor/
MIT License
53 stars 24 forks source link

Exposure issues with OV5647 #53

Closed spamekkel closed 3 weeks ago

spamekkel commented 3 weeks ago

Thanks for the excellent tutorials and examples, very much appreciated!

After flashing the YoloV8 OD model application with the OV5647 image sensor, I noticed the image is very much underexposed when running the application on a Grove Vision AI V2 board. The driver is found in HIMAX_WE2_Rabboni/EPII_CM55M_APP_S/app/scenario_app/tflm_yolov8_od/cis_sensor/cis_ov5647 and uses the OV5647_mipi_2lane_640x480.i configuration.

I'm using the https://sensecraft.seeed.cc/ai/#/device/.... webpage to display the result. Here I only get a singhle image from the sensor, perhaps the Baudrate is too high to catch up with the JSON string formated image.

Until now I have not found the reason. Any clue, somebody?

kris-himax commented 3 weeks ago

Hi @spamekkel

Did you try https://github.com/HimaxWiseEyePlus/Seeed_Grove_Vision_AI_Module_V2/tree/main/EPII_CM55M_APP_S/app/scenario_app/tflm_yolov8_od this repository and use Himax AI tool to show? The readme about Himax AI tool you can reference here.

And make sure that your code has added the hx_InvalidateDCache_by_Addr() before you send the jpeg image by UART.

Thanks, Kris

spamekkel commented 3 weeks ago

Hi Kris,

I just found the cause: the model was not at the correct flash address. Thanks for your help.

Jan.