5 ### Compiler supports more operations
7 - NonMaxSuppressionV4, NonMaxSuppressionV5, PadV2, Unique
11 - Quantization enhancements: channel wise UINT8 quantization(Conv2D, DepwiseConv, TransposeConv, FullyConnected)
12 - Experimental requantization from INT8 to UINT8
13 - Adding more operator value tests
14 - tf2tfliteV2 supports conversion from Keras model, saved model
15 - Refactoring for better maintenance long Class codes using visitor patterns
16 - Introducing optimization pass that fuses batch normalization with Transposed Convolution.
21 ### Runtime backend operation support
24 - CPU backend qasymm uint8: LOG_SOFTMAX
25 - ACL-CL backend: LEAKY_RELU, RESIZE_NEAREST_NEIGHBOR
30 - Copy Elimination between compatible backends
32 ### Operation Implementation
34 - Operations with same parameters are unified
38 - CPU backend qasymm uint8 performance enhancement: arithmetic operations