Open JP-devv opened 1 year ago
Hi there! well, I made a synthetic dataset and trained the model. After that, I quantized the model and converted to tflite file as explained in the Readme file. Then you can put the tflite file in the bitcraze virtual machine and generate the AT model (this is done using the commands we provide). You can call the generated header files in your C based application code.
Lately, the most relevant obstacle that I encountered was that TensorFlow doesn't quantize properly for GAP8 architecture, so I have been quantizing my models directly in the virtual machine.
Best!
Angel Canelo Hello I found your example insect inspired image recognition cnn by chance and I am very interested in trying to use in my Crazyflie, what I want to ask is if you need to have a python script to be able to run it. Regards Rodrigo Calvo Email: rudiger46@yahoo.com.au
Hi there! In our project the code files are flashed in the drone and AI deck memories so we didn't need any python script. We only used python to develop our neural network model using tensorflow and convert it to TFlite file. All details are in the readme file step by step. Regards
hello thank you for your answer. I have flashed the code on the crazyflie, now i need some directions on how to actually run the code. i am not sure what the next step is. Thank you regards
hello thank you for your answer. I have flashed the code on the crazyflie, now i need some directions on how to actually run the code. i am not sure what the next step is. i have read the documentation several times and i cant figure out the next step. Thank you regards rodrigo calvo Email: rudiger46@yahoo.com.au
Hi i'm developing a similar project with the crazy flie, except it will be recognizing humans from an aerial POV. What did you use to train the CNN model and how were you able to integrate it to the AI deck? What obstacles did you encounter if any?