标题: Not surpport tensorflow version of yolov4-tiny [打印本页] 作者: elexxy 时间: 2020-8-26 09:28 标题: Not surpport tensorflow version of yolov4-tiny 本帖最后由 elexxy 于 2020-8-26 09:31 编辑
E Try match Split split_2 failed, catch exception!
W ----------------Warning(2)----------------
E Catch exception when loading tensorflow model: yolov4-tiny/yolov4-tiny.pb!
E Traceback (most recent call last):
E File "rknn/base/RKNNlib/converter/convert_tf.py", line 713, in rknn.base.RKNNlib.converter.convert_tf.convert_tf.match_paragraph_and_param
E AttributeError: 'Split' object has no attribute 'load_params_from_tf'
E During handling of the above exception, another exception occurred:
E Traceback (most recent call last):
E File "rknn/api/rknn_base.py", line 374, in rknn.api.rknn_base.RKNNBase.load_tensorflow
E File "rknn/base/RKNNlib/converter/convert_tf.py", line 767, in rknn.base.RKNNlib.converter.convert_tf.convert_tf.match_paragraph_and_param
E File "rknn/api/rknn_log.py", line 312, in rknn.api.rknn_log.RKNNLog.e
E ValueError: Try match Split split_2 failed, catch exception!
For some reason I have to use one channel images as inputs, but the rknn toolkit does not support one channel darknet model's quantization.So I convert the darknet model to a tensorflow model and it works well before.However, today I try the yolov4 tiny model in this way, and the rknn toolkit return above error.It seems like dose not support the tf.split function.I hope you guys can support more functions and notice the problem of one channel darknet model's quantization.Tanks.
作者: elexxy 时间: 2020-8-26 09:34
PS. I have try it on 1.3.2 and 1.4.0 version, and it dose not work.作者: jefferyzhang 时间: 2020-8-26 14:44
NPU Team repeat:
We have successfully converted darknet offical edition yolov4-tiny (3-channels version) to rknn with rknn-toolkit 1.4.0.
Please try it. And your LOG looks can not support the op "split_2" in tf-edittion.