Huawei uses machine translation combined with human proofreading to translate this document to different languages in order to help you better understand the content of this document.
Note: Even the most advanced machine translation cannot match the quality of professional translators.
Huawei shall not bear any responsibility for translation accuracy and it is recommended that you refer to the English document (a link for which has been provided).
Tensor Decomposition Specification Reference of Open-Source Networks
The accuracy specification is top1 ACC(%) for an image classification network, mAP(%) for an object detection network, or DSC(%) for an image segmentation network. The fine-tuning learning rate decreases from 0.1 times of the original learning rate.
Network |
Job Type |
Dataset |
Baseline Accuracy |
Accuracy After Decomposition |
Accuracy After Post-decomposition Fine-tuning |
|---|---|---|---|---|---|
ResNet-18 |
Classification |
ImageNet |
70.66 |
44.02 |
70.34 |
ResNet-34 |
Classification |
ImageNet |
74.2 |
54.92 |
74.15 |
ResNet-50 |
Classification |
ImageNet |
75.6 |
73.64 |
75.91 |
ResNet-101 |
Classification |
ImageNet |
78.52 |
76.97 |
78.24 |
InceptionV3 |
Classification |
ImageNet |
77.98 |
76.95 |
77.78 |
SSD |
Detection |
COCO 2017 |
27.2 |
24.2 |
27.9 |
Faster R-CNN |
Detection |
COCO 2017 |
32.5 |
31 |
32.2 |
Mask R-CNN |
Detection |
COCO 2017 |
37.9 |
36.8 |
38 |
UNet |
Segmentation |
SSTEM |
87.63 |
85.05 |
87.57 |
