郑启航
郑启航
k210是通过kpu对模型进行加速的, 所以只有kpu支持的模型才可以在k230上流畅运行.你可以参考nncase项目的faq. 你可以先去学习一下吴恩达的深度学习课程, 然后再学习一些关于模型部署的知识.
自定义数据集的时候很容易出现数据太少,并且我之前的代码是对于测试集是直接取训练集中的一部分,可能导致他无法满足`tf.dataset`里面的buffer出现这个问题,你可以在定义输入数据管道的地方将`shuffle`、`map`等操作的buffer改小一些。
https://github.com/zhen8838/K210_Yolo_framework/tree/master/yolo3_frame_test_public_maixpy 这个我在maix go上测试过,应该和maix dock通用.
yes , you can. but mobilenet backbone for k210 need to modify. you can refer my code `https://github.com/zhen8838/K210_Yolo_framework/blob/master/models/keras_mobilenet.py`
This error does not affect the model freeze
@yakeer I have the same error message here, but I have tested that the generated tflite model does not have a loss of precision.
在Uubuntu开发的,window应该也支持。把自己的数据集整理成voc格式,然后用脚本重新生成一下就好了。
用老版本的nncase进行量化
0.2fps may be because you didn't rewrite `region layer.c`. the real-time demo detection code I showed was rewritten by me. The output is the same, You should refer to the...
demo link: `https://github.com/kendryte/nncase/tree/master/examples/20classes_yolo`