理论计算时间(单图) | 1.79ms | 8.23ms | 7.485ms | 30.55ms |
实际计算时间(单图) | 21.37ms | 39.82ms | 33.12ms | 76.13ms |
- export NN_LAYER_DUMP=1
- python xxx.py
复制代码
- layer {
- name: "data"
- type: "Input"
- top: "data"
- input_param {
- shape {
- dim: 1
- dim: 3
- dim: 224
- dim: 224
- }
- }
- }
复制代码
3.2.转换模型时,出现“Message type "caffe.PoolingParameter" has no fieldnamed "round_mode"”错误。T raise ValueError("'%s' is not a valid scope name" % name)T ValueError: '_plus0_17' is not a valid scope name
- [A-Za-z0-9.][A-Za-z0-9_.\\-/]* (for scopes at the root)
- [A-Za-z0-9_.\\-/]* (for other scopes)
复制代码
- layer {
- name: 'proposal'
- type: 'proposal'
- bottom: 'rpn_cls_prob_reshape'
- bottom: 'rpn_bbox_pred'
- top: 'rois'
- top: 'scores'
- proposal_param {
- ratio: 0.5 ratio: 1.0 ratio: 2.0
- scale: 8 scale: 16 scale: 32
- base_size: 16
- feat_stride: 16
- pre_nms_topn: 6000
- post_nms_topn: 300
- nms_thresh: 0.7
- min_size: 16
- }
- }
复制代码
- rknn.load_tensorflow(tf_pb='./ssd_mobilenet_v2_coco_2018_03_29/frozen_inference_graph.pb',
- inputs=['FeatureExtractor/MobilenetV2/MobilenetV2/input'],
- outputs=['concat', 'concat_1'],
- input_size_list=[[INPUT_SIZE, INPUT_SIZE, 3]])
复制代码
momo 发表于 2019-4-28 16:52
楼主,关于dataset的图片,采用模型训练时用的图片吗?
vanilla 发表于 2019-5-6 11:06
请教楼主,还不是很懂量化时为啥要做inference,直接用model里的各层权值不能计算出zero point这些数值吗? ...
metaphor22 发表于 2019-5-7 09:31
请问楼主,rknn的源码开放吗?是否可以通过商业途径向你们获取?
欢迎光临 Toybrick (https://t.rock-chips.com/) | Powered by Discuz! X3.3 |