|
Chris Pollett >
Students >
Deshmukh
( Print View)
[Bio]
[Blog]
[CS 297 Proposal]
[Deliverable 1]
[Deliverable 2]
[Deliverable 3]
[Deliverable 4]
[End to End Image Compression - PDF]
[Image file formats - PDF]
[DeepN-JPEG - PDF]
[Image compression using RNN - PDF]
[CS 297 Report - PDF]
[CS 298 Proposal]
[CS 298 report]
|
Project Blog
What I found Interesting this week
Meeting 26: March 24, 2019
Presented first version of paper
Tasks assigned:
- Remove "1st" and "2nd" before author names
- Use only your own diagrams.
- Perform experiments on various quality factors for JPEG across bits per pixel values.
GRADUATION
Meeting 17: March 19, 2019
Modular code works. Presented findings about performance of a loss function
Tasks assigned:
- Complete model code with all layers and perform hyper-parameter tuning
- Generate more data with data augmentation
- list down all the experiments performed on the model.
Meeting 16: March 12, 2019
Still working on a modular code.
Tasks assigned:
- Get the modular code running
- Perform experiments on other version - loss function and RNN layer for encoding
Meeting 15: March 05, 2019
Discussed problem with reconstructed image. Image quality is bad enevn though PSNR and MSE metrics are good.
Tasks assigned:
- Improve compressed and output image quality for human eye.
- Write "Related work" section.
Meeting 14: February 26, 2019
Discussed neural network structure, loss function
Tasks assigned:
- Implement the changes in neural network as discussed.
- Write "Related work" section.
Meeting 13: February 19, 2019
Discussed about Dataset to be used in the project, program structure
Tasks assigned:
- Develop and train a baseline DNN architecture for image compression
- Report : Related work
Meeting 12: February 12, 2019
presented a code that splits images in 'n' subplots
Tasks assigned:
- Develop an architecture including training algorithm.
- Can we treat the important features in an image differently?
CS 298
Meeting 11: December 04, 2018
Presented first draft of CS297 report, CS 298 proposal
Tasks assigned:
- Make changes to the CS297 report as suggested and upload it on website after second review.
Meeting 10: November 27, 2018
Presented findings from Deliverable 1 & 3, Discussed a paper on image compression using recurrent neural network
Tasks assigned:
- Work on Deliverable - 4
- Complete CS 297 repot
- Complete CS 298 proposal
Meeting 9: November 13, 2018
Presented findings from Deliverable 1,2 & 3
Tasks assigned:
- Calculate matrics such as MSE, SSIM on deliverable-2 & 3
- Implement GAN on better image database
- Work on Deliverable - 4
Meeting 8: October 23, 2018
Discussed problems with Deliverable - 1,2. Discussed deliverable 3
Tasks assigned:
- Work on Deliverable - 2
- Deliverable 1,3.
Meeting 7: October 16, 2018
Discussed problems with Deliverable - 1. Discussed on a papers - Deep Neural Network Favorable JPEG-based Image Compression Framework and Joint Autoregressive and Hierarchical Priors for Learned Image Compression
Tasks assigned:
- Work on Deliverable - 2
- Upload the slides used so far.
- Upload Deliverable 1 report
- Read a paper : Joint Autoregressive and Hierarchical Priors for Learned Image Compression arXiv:1809.02736v1
Meeting 6: October 9, 2018
Image encoding formats and Errors in Deliverable-1 discussed. Professor explained Pixelshuffler layer.
Tasks assigned:
- Implement a research paper on Photo realistic Super resolution using GAN
- Upload the slides used so far.
- Work on Deliverable 1.
- Read a paper : Joint Autoregressive and Hierarchical Priors for Learned Image Compression arXiv:1809.02736v1
- Read a paper : DeepN-JPEG: A Deep Neural Network Favorable JPEG-based Image Compression Framework
Meeting 5: October 2, 2018
End to End framework for image compression tested. Deliverable 1 complete. paper on photo-relistic Single Image Super-Resolution discussed.
Tasks assigned:
- Implement a research paper on Photo realistic Super resolution using GAN
- Study image encoding formats with better compression ratio than JPEG
- Add a report on Deliverable 1 and its source code to this site.
- [optional] Read a paper : Joint Autoregressive and Hierarchical Priors for Learned Image Compression arXiv:1809.02736v1
Meeting 4: September 25, 2018
End to End framework for image compression training completed.
Tasks assigned:
- Represent intermediate and final result of a network in the for of JPEG image
- Study image encoding formats with better compression ratio than JPEG
- Read a paper : Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
- [optional] Read a paper : Joint Autoregressive and Hierarchical Priors for Learned Image Compression arXiv:1809.02736v1
- [optional] Start working on Deliverable 2
What i found interesting this week :
- Free stock photos : https://www.pexels.com
Meeting 3: September 18, 2018
Discussed on loss function of "An End-to-End Compression Framework Based on Convolutional Neural Networks" and how its code can be executed in tensorflow
Tasks assigned:
- Complete the code for "An End-to-End Compression Framework Based on Convolutional Neural Networks"
What i found interesting this week :
- Joint Autoregressive and Hierarchical Priors for Learned Image Compression arXiv:1809.02736v1
Meeting 2: September 11, 2018
Discussed about 297 project proposal reference format.
Presented amended project proposal and discussed a research paper - "An End-to-End Compression Framework Based on Convolutional Neural Networks"
Tasks assigned:
- Change references in project proposal to APA format and cite references wherever appropriate.
- Read about article on JPEG published in ACM communication
- Write a code to implement CNN framework suggested in "An End-to-End Compression Framework Based on Convolutional Neural Networks"
Meeting 1: September 04, 2018
Presented a project proposal
Tasks assigned:
- Make changes in project proposal as advised.
- Read a research paper - "An End-to-End Compression Framework Based on Convolutional Neural Networks"
CS 297
|