draw

KO | EN

gitlite — search

#javascript #react #llm #go #machine-learning #rust #linux #macos #windows #awesome #awesome-list #security #ios #open-source

(★ 136)

Torch implementation of DRAW: A Recurrent Neural Network For Image Generation

File Explorer

Download Latest Version (.zip)

- Auto-Encoding Variational Bayes.pdf
- Deep AutoRegressive Networks.pdf
- DRAW.pdf
.gitignore
DRAW.pdf
draw_attention.lua
draw_attention_read.lua
draw_no_attention_gaussian.lua
draw_no_attention_sigmoid.lua
model_utils.lua
Plot_result_no_binarization_example.png
plot_results.lua
Plot_results_example.png
plot_results_no_binarization.lua
read_patch.lua
README.md
selective.lua
te.lua
te1.lua
te2.lua
write_patch.lua

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

Torch implementation of DRAW: A Recurrent Neural Network For Image Generation http://arxiv.org/pdf/1502.04623.pdf. Watch Deep Learning Lecture 14: Karol Gregor on Variational Autoencoders and Image Generation https://www.youtube.com/watch?v=P78QYjWh5sM&list=PLE6Wd9FR--EfW8dtjAuPoTuPcqmOV53Fu&index=3

Run 
```th draw_attention.lua```
in Terminal.app, it generates `x_prediction`, which you can plot by running `plot_results*.lua` in zbs-torch (https://github.com/soumith/zbs-torch) with QLua-LuaJit interpreter selected from 'Project' tab. Adjust the running time of the script by changing:
```
1. n_data (the number of MNIST examples to train on)
2. number of iterations
3. n_z, dimension of the hidden layer z
4. rnn_size, dimension of h_dec and h_enc
```

draw_attention.lua works with 28x28 MNIST dataset. You can adjust it to other datasets by changing A, N and replacing number '28' everywhere in the script. I haven't done it but it is possible.

draw_no_attention*.lua scripts implement DRAW without attention.
In draw_attention_read.lua only read is attentive, while write is without attention.

draw_no_attention*.lua scripts print arrays in the end, which helps to quickly estimate the quality of the results without plotting

Example output by plot_results.lua
![th visualize_word_vectors.lua](https://github.com/vivanov879/draw/blob/master/Plot_results_example.png)

Example output by plot_results_no_binarization.lua
![th visualize_word_vectors.lua](https://github.com/vivanov879/draw/blob/master/Plot_result_no_binarization_example.png)

draw

File Explorer

Today's picks

Was this content helpful?

Recent Feedback

draw

File Explorer

Today's picks

Was this content helpful?

Recent Feedback

Download README