/jpeg_tcm

Primary LanguageC++

JPEG with Transparent Composite Model (TCM)

Authors: Hossam Amer (@HossamAmer12) and Yanbing Jiang (@YanbingJiang)

Created: September 2018

Content Included

  • JPEG Baseline Model Decoder and Encoder with Implementation of TCM

Prerequisites

  • Enviornment: C++
  • IDE: Visual Studio 2017 (Windows), XCode (MAC OS)

Updates

  • 2018-09-14: First Upload, Sequential Model Done (0xFFC0)
  • 2018-09-17: Progressive: Fifth Scan Done for goose_org.jpeg, Huffman Table overwritten, Dequantiz not done
  • 2018-09-20: Progressive: All Scan Done for goose_org.jpeg, Huffman Table overwritten, Dequantiz not done
  • 2018-09-23: Progressive: All Scan done, Data Read properly, but one tcoeff reason unknown, Huffman Table overwritten
  • 2018-09-24: Tested on Multiple SOI (0xFFD8): Success, Progressive Decoder Clean up, Y Component Shown in Release Mode of VS2017 but not Debug Mode
  • 2018-09-25: RGB-BMP Picture shown bug fixed
  • 2018-09-26: Progressive Encoder Standard Default Huffman Table Built
  • 2018-09-26: Support multiple compilation modes and added code for Huffman tables and SOF writing in the progressive mode (trying to write as sequential)
  • 2018-09-27: Progressive Encoder as Sequential with Standard Default Huffman Table Built Done
  • 2018-09-28: Support Multiple picture processing (decode and encode), utilize with flag "MULTI_PIC"
  • 2018-10-01: Multion SOI (0xFFD8) bug found and fixed, not dog.JPEG/lion_org.jpg can be processed
  • 2018-10-10: Support adjust Picture with Quality Factor (var QFACTOR in Typedef.h, not tested due to VS2017 issue)
  • 2018-10-15: Default huffman table writing is done to QF Experiement and Same QTable for QTable experiment is done
  • 2018-10-16: QF Experiment Configuration for VS2017 is done using default QT and HT as standard
  • 2018-10-17: Rounding value has to exist in Quantization process and Fix default HT
  • 2018-10-20: QF Experiment Configuration for dataset finalized and fix BW picture bug
  • 2018-10-25: Added Inception V3 Group Running Python Sheet and MATLAB analysis sheet
  • 2018-11-04: Optimised the JPEG decoder and the encoder to run 4k Pictures in about 20 seconds
  • 2018-11-04: Parallelized the quality factor execution and updated the bash script to account for this parallelism
  • 2018-11-05: Added Inception ResNet V2 Group Running Python Sheet and optimised the JPEG decoder and the encoder for VS
  • 2018-11-23: FC stage begin: Added the neural network folder + fixed bugs with CMYK and non-parsebale pictures
  • 2018-12-03: FC: Model getting 92 Vs 91... learning rate after 150 decrease by 0.1... early stopping implemented one layer
  • 2018-12-04: FC: Base FC Model Generated and realize parallel running in QF's bottleneck collection
  • 2018-12-05: FC: Adding the predict for new architecture (WIP)
  • 2018-12-18: FC: Commit the latest Bottleneck generation Python script for efficiency purpose
  • 2019-01-07: ImageNet Training Images Bottlenect generated with retrain_attempt_imagenet_v3.py
  • 2019-01-15: FC: Re-train architecture built and ready to retrain
  • 2019-01-22: ImageNet Validation Data Bottlenect generated with retrain_attempt_imagenet_v3.py