Hyukjin Jeong

## What Let's support quantization of hybrid FullyConnected Op (int8 weights/float32 input -> uint8 weights/uint8 input or int16 weights/int16 input). Parent issue: https://github.com/Samsung/ONE/issues/9535

## What Let's support quantization of (TF-style) hybrid Ops whose input feature map is float32 and weights(filters) are quantized. List of hybrid Ops - FullyConnected - Conv2D - DepthwiseConv2D -...

[Compiler FE] Convert dynamic batch to single batch

2

## What Let's implement an option to convert a model with a dynamic batch size to a model with a single batch. For example, ``` onecc optimize --dynamic_batch_to_single_batch -i dynamic.circle...

[circle-optimizer] Support constant folding of Gather/SparseToDense

4

## What Let's support constant folding of Gather and SparseToDense. ## Why Reshape Op in onnx model is translated to Gather->SparseToDense->AddV2->Cast in circle. onnx  circle  A similar pattern...

[circle2circle] Enable proper options when converting NCHW to NHWC

1

This automatically enables proper options when convertin NCHW to NHWC. ONE-DCO-1.0-Signed-off-by: Hyukjin Jeong --- Related to #7350

This introduces visq for visualizing quantization error. ONE-DCO-1.0-Signed-off-by: Hyukjin Jeong --- Related to: https://github.com/Samsung/ONE/issues/9694 Draft PR: #9868

DRAFT: Introduce visq module

6

On going draft to introduce visq module. ONE-DCO-1.0-Signed-off-by: Hyukjin Jeong --- Related to: https://github.com/Samsung/ONE/issues/9694

On going draft to introduce ForwardTransposeOpPass. ONE-DCO-1.0-Signed-off-by: Hyukjin Jeong --- Related to: https://github.com/Samsung/ONE/issues/9872

[one-optmize] Introduce ForwardTransposeOpPass

1

## What Let's introduce `ForwardTranseposeOpPass`. ## Why This moves Transpose Ops to the rear part, finding a chance to optimize redundant Transpose Ops (similar to `ForwardReshapeToUnaryOpPass`). Before  After 

[quantizer] Visualize layer-wise qerror

25

## What Let's visualize quantization error in the quantized model ## Why To observe which quantized layer harms accuracy. ## How Input: fp32 circle, q circle, test data Steps 1....