Conversation
| | Models | Accuracy(%) | | ||
| | ------ | -------------- | | ||
| | CRNN_EN | 81.66 | | ||
| | CRNN_EN_FP16 | 82.01 | | ||
| | CRNN_CH | 71.28 | | ||
| | CRNN_CH_FP16 | 78.63 | | ||
| \*: 'FP16' stands for 'model quantized into FP16'. |
There was a problem hiding this comment.
We should carry experiment on more OCR datasets. Then, I think we can merge it. Please add info about the test dataset.
|
We need to add notes in the readme telling users how to use these fp16 models and compare the accuracy with fp32 models' as well. |
|
@fengyuentau This info can be found in the merged PR #71. |
|
Thank you but we still need notes in the readme about how to use these fp16 models with opencv. |
|
In my opinion, we will have more fp16 models in the future, we need to add a unified note about how to load float16 in OpenCV. How about adding this info in https://github.com/opencv/opencv_zoo/tree/master/benchmark/README.md? |
* add crnn CH FP16 model * add crnn EN FP16 model
add two crnn quantized models