[GSoC] Add block quantized models (#270)

* Gemm and MatMul block quantization support

* refactoring

* fix indentation

* node name independent

* Block quantization tool:
- constant weight category supported
- add data type saturation
- handled the case in which all the elements within a block are the same

benchmark script modified to support block quantized models

block quantized some models

* add missing block quantized models

* formatting

* add blocked models to eval script. Evaluation yunet

* Add sface and pphumanseg evaluation, block quantization tool fix, handpose blocked model fix, removed blocked CRNN EN,

* changed evaluation metric in block_quantize script and add verbose mode

* Add evaluation for PP-ResNet and Mobilenet

* changed file suffix and update readmes

* renamed int8bq

README.md

@@ -8,15 +8,18 @@ Note:
 - Model files encode MobileFaceNet instances trained on the SFace loss function, see the [SFace paper](https://arxiv.org/abs/2205.12010) for reference.
 - ONNX file conversions from [original code base](https://github.com/zhongyy/SFace) thanks to [Chengrui Wang](https://github.com/crywang).
 - (As of Sep 2021) Supporting 5-landmark warping for now, see below for details.
+- `face_recognition_sface_2021dec_int8bq.onnx` represents the block-quantized version in int8 precision and is generated using [block_quantize.py](../../tools/quantize/block_quantize.py) with `block_size=64`.
 
 Results of accuracy evaluation with [tools/eval](../../tools/eval).
 
 | Models      | Accuracy |
 | ----------- | -------- |
 | SFace       | 0.9940   |
+| SFace block | 0.9942   |
 | SFace quant | 0.9932   |
 
 \*: 'quant' stands for 'quantized'.
+\*\*: 'block' stands for 'blockwise quantized'.
 
 ## Demo