人工智能开发系列(3) YOLOV3开发与实现

TryYourBest 发表于 2019-6-27 09:42
在跑案例的过程中，运行“python3 rknn_transform.py”时，当进行到“Building model1”时，经过一段时间后 ...

应该是板子内存不足，我也遇到了，然后在pc上转换成功了

能不能把源代码公开一下，让大家做一下参考。

我再1.2.1版本上跑yolov3，模型加载很快能完成，但第一次推理速度极慢，是怎么回事

请问楼主的rknn-toolkit是哪个版本？ 我在1.0.0和1.2.1版本上，python3 rknn_transform_320x320.py 出现错误：
--> Loading model
E Catch exception when loading darknet model: ./yolov3_320x320.cfg!
E Traceback (most recent call last):
E   File "rknn/api/rknn_base.py", line 380, in rknn.api.rknn_base.RKNNBase.load_darknet
E   File "rknn/base/RKNNlib/converter/convert_darknet.py", line 471, in rknn.base.RKNNlib.converter.convert_darknet.Converter.load_blobs
E   File "rknn/base/RKNNlib/converter/convert_darknet.py", line 215, in rknn.base.RKNNlib.converter.convert_darknet.Convolution_Blob_Assign.assign
E   File "<__array_function__ internals>", line 6, in reshape
E   File "/home/abc/.local/lib/python3.6/site-packages/numpy/core/fromnumeric.py", line 301, in reshape
E     return _wrapfunc(a, 'reshape', newshape, order=order)
E   File "/home/abc/.local/lib/python3.6/site-packages/numpy/core/fromnumeric.py", line 61, in _wrapfunc
E     return bound(*args, **kwds)
E ValueError: cannot reshape array of size 148814 into shape (256,128,3,3)
done

appleUncle 发表于 2019-12-25 10:22
请问楼主的rknn-toolkit是哪个版本？ 我在1.0.0和1.2.1版本上，python3 rknn_transform_320x320.py 出现错 ...

自己解决了，是yolo的wieghts下载不全，重新下载就好了
cd yolov3_demo
wget https://pjreddie.com/media/files/yolov3-tiny.weights
wget https://pjreddie.com/media/files/yolov3.weights

楼主，我运行tiny没问题，但为什么运行于320,416和608的时候卡在init_runtime之后就显示显示内存超出

tiny版的和v3版的转换和测试程序一样吗？除了v3版多一个52那层，还有别的地方需要更改吗？为什么跑测试，v3的结果就是正确的，tiny版总有很多重复的框，且框的位置也不准，和原darknet的测试结果不同。。

转换tiny的时候会报以下错误

$ python rknn_transform_tiny.py
--> Loading model
E Catch exception when loading darknet model: ./yolov3-tiny.cfg!
E Traceback (most recent call last):
E   File "rknn/api/rknn_base.py", line 511, in rknn.api.rknn_base.RKNNBase.load_darknet
E   File "rknn/base/RKNNlib/converter/convert_darknet.py", line 685, in rknn.base.RKNNlib.converter.convert_darknet.Converter.load_blobs
E   File "rknn/base/RKNNlib/converter/convert_darknet.py", line 403, in rknn.base.RKNNlib.converter.convert_darknet.Convolution_Blob_Assign.assign
E   File "/home/nationalflag/.local/lib/python3.6/site-packages/numpy/core/fromnumeric.py", line 292, in reshape
E     return _wrapfunc(a, 'reshape', newshape, order=order)
E   File "/home/nationalflag/.local/lib/python3.6/site-packages/numpy/core/fromnumeric.py", line 56, in _wrapfunc
E     return getattr(obj, method)(*args, **kwds)
E ValueError: cannot reshape array of size 3650443 into shape (1024,512,3,3)
done
--> Building model
Traceback (most recent call last):
  File "rknn_transform_tiny.py", line 27, in <module>
    rknn.build(do_quantization=True, dataset='./dataset_416x416.txt')
  File "/home/nationalflag/.local/lib/python3.6/site-packages/rknn/api/rknn.py", line 220, in build
    inputs = self.rknn_base.net.get_input_layers()
AttributeError: 'NoneType' object has no attribute 'get_input_layers'

训练的时候loss一直是nan 训练几次IOU也变成了NULL   rate出奇的高
Region 16 Avg IOU: 0.443191, Class: 0.426055, Obj: 0.426348, No Obj: 0.442097, .5R: 0.500000, .75R: 0.000000,  count: 4
Region 23 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.522350, .5R: -nan, .75R: -nan,  count: 0
Region 16 Avg IOU: 0.543481, Class: 0.443554, Obj: 0.483441, No Obj: 0.442328, .5R: 0.800000, .75R: 0.000000,  count: 5
Region 23 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.521949, .5R: -nan, .75R: -nan,  count: 0

1: 748.069458, 748.069458 avg loss, 0.000000 rate, 457.962847 seconds, 64 images
Loaded: 0.000054 seconds
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: -nan, .5R: 0.000000, .75R: 0.000000,  count: 6
Region 23 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: -nan, .5R: -nan, .75R: -nan,  count: 0
Region 16 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: -nan, .5R: 0.000000, .75R: 0.000000,  count: 11
刚开始还会报以下错误
Can't open label file. (This can be normal only if you use MSCOCO): /home/gec/ros/test1_ws/src/obj_detect/src/darknet/train_my_data/train_images/banana_70.txt
Can't open label file. (This can be normal only if you use MSCOCO): /home/gec/ros/test1_ws/src/obj_detect/src/darknet/train_my_data/train_images/apple_81.txt
Can't open label file. (This can be normal only if you use MSCOCO): /home/gec/ros/test1_ws/src/obj_detect/src/darknet/train_my_data/train_images/apple_81.txt
Can't open label file. (This can be normal only if you use MSCOCO):

你好，在PC上执行python3 rknn_transform_320x320.py会出现
W The RKNN Model generated can not run on simulator when pre_compile is True.
然后在板子上跑程序会出现初始化失败的错误，ret=-6
我在PC上跑python3 rknn_picture_320x320.py也会报错
-->loading model
loading model done
--> Init runtime environment
E Pre compile model can not run on simulator
E Catch exception when init runtime!
E Traceback (most recent call last):
E   File "rknn/api/rknn_base.py", line 988, in rknn.api.rknn_base.RKNNBase.init_runtime
E   File "rknn/api/rknn_runtime.py", line 287, in rknn.api.rknn_runtime.RKNNRuntime.build_graph
E Exception: RKNN init failed. Wrong platform: simulator
Init runtime environment failed