Deval's daily progress report

Day 1 (22nd May 2018)

• Researched about types of GANs
• Looked up ways to generate images from some text by creating a text encoding
• Researched ways to generate video from images
  • https://arxiv.org/pdf/1605.05396.pdf
  • https://arxiv.org/pdf/1704.00028.pdf
  • https://arxiv.org/pdf/1411.1784.pdf

---

Day 2 (23rd May 2018)

• Found a way to use an LSTM network to generate video from images
• Generating videos from images by separating the static background from foreground
• Referred some PyTorch tutorials (refer)
• Started to work on the report as to how to tackle the problem, various algorithms available
  • https://arxiv.org/pdf/1605.08104
  • http://carlvondrick.com/tinyvideo/paper.pdf

---

Day 3 (25th May 2018)

• Completed the task 1 report
• Read about temporal conditioned GANs to generate video from text
• Discussed Task 2 details
  • https://arxiv.org/pdf/1611.06624.pdf
  • https://www.microsoft.com/en-us/research/wp-content/uploads/2017/11/BNI02-panA.pdf

---

Day 4 (26th May 2018)

• Went through a few GAN Tutorials
• Went through a few pyTorch Tutorials
• Looked up DCGAN architecture details
  • https://medium.com/ai-society/gans-from-scratch-1-a-deep-introduction-with-code-in-pytorch-and-tensorflow cb03cdcdba0f
  • https://github.com/pytorch/examples/tree/master/dcgan
  • https://github.com/znxlwm/pytorch-MNIST-CelebA-GAN-DCGAN

---

Day 5 (28th May 2018)

• Programmed the code for DCGAN
• Trained the DCGAN model

---

Day 6 (29th May 2018)

• committed task 2 with comments
• discussed task 2 details
• researched more about text-to-image
  • https://arxiv.org/abs/1612.04357
  • https://arxiv.org/abs/1605.05396
  • https://github.com/reedscot/icml2016

---

day 7 (31th May 2018)

• started work on creating the model
• found out about stack GAN
• looked up datasets for training our model
  • https://cocodataset.org
  • https://github.com/hanzhanggit/StackGAN-Pytorch
  • https://github.com/aelnouby/Text-to-Image-Synthesis/blob/master/utils.py

day 8 (1st June 2018)

• still working on creating the model
• looking up datasets for text captions
  • https://github.com/reedscott/cvpr2016

day 9 (2nd June 2018)

• started work on the progress presentation
• dataset generation script was prepared

day 10(6th June 2018)

• finalized the ppt for now
• removed last of the errors from the model

day 11(7th June 2018)

• put the model on training.
• progress presentation performed.

day 12(8th June 2018)

• found out that the training was unsuccessful as the images were very poor quality and the loss curve was flat lining
• made some improvements to the model.

day 13(9th June 2018)

• the training images were yet the same and the loss curve continued to flat line
• decided to change the model and use stackGAN.

day 14(11th June 2018)

• the training had completed and some images were generated from the coco dataset.
• looking to find appropriate dataset for video generation. -- http://www.nada.kth.se/cvap/actions/ -- http://crcv.ucf.edu/data/UCF101.php -- http://www.cs.toronto.edu/~nitish/unsupervised_video/

day 15(12th june 2018)

• as per kalind sir's suggestion used the birds/flowers dataset for image generation.
• prepared the dataloader for birds dataset. -- http://www.vision.caltech.edu/visipedia/CUB-200-2011.html -- http://www.robots.ox.ac.uk/~vgg/data/flowers/102/

day 16(13th June 2018)

• put the flowers model for training.
• continued work for video dataset and dataloader.

day 17(14th June 2018)

• put the birds model on training