cnn-toys
所属分类：工具库
开发工具：Python
文件大小：0KB
下载次数：0
上传日期：2018-09-17 23:11:51
上 传 者：
sh-1993
说明： 玩CNN
(Playing around with CNNs)
文件列表:
cnn_toys/ (0, 2018-09-17)
cnn_toys/__init__.py (0, 2018-09-17)
cnn_toys/colorize/ (0, 2018-09-17)
cnn_toys/colorize/model.py (1759, 2018-09-17)
cnn_toys/colorize/run_sample.py (1404, 2018-09-17)
cnn_toys/colorize/run_train.py (1841, 2018-09-17)
cnn_toys/cyclegan/ (0, 2018-09-17)
cnn_toys/cyclegan/__init__.py (194, 2018-09-17)
cnn_toys/cyclegan/history.py (1829, 2018-09-17)
cnn_toys/cyclegan/model.py (7550, 2018-09-17)
cnn_toys/cyclegan/run_single.py (1798, 2018-09-17)
cnn_toys/cyclegan/run_train.py (3844, 2018-09-17)
cnn_toys/cyclegan/test_history.py (2068, 2018-09-17)
cnn_toys/data.py (2336, 2018-09-17)
cnn_toys/graphics.py (1285, 2018-09-17)
cnn_toys/real_nvp/ (0, 2018-09-17)
cnn_toys/real_nvp/__init__.py (448, 2018-09-17)
cnn_toys/real_nvp/interp.py (1105, 2018-09-17)
cnn_toys/real_nvp/layer.py (19665, 2018-09-17)
cnn_toys/real_nvp/models.py (717, 2018-09-17)
cnn_toys/real_nvp/objective.py (2440, 2018-09-17)
cnn_toys/real_nvp/run_interp.py (2018, 2018-09-17)
cnn_toys/real_nvp/run_sample.py (1916, 2018-09-17)
cnn_toys/real_nvp/run_train.py (3022, 2018-09-17)
cnn_toys/real_nvp/test_layer.py (6215, 2018-09-17)
cnn_toys/saving.py (809, 2018-09-17)
cnn_toys/schedules.py (581, 2018-09-17)
setup.py (348, 2018-09-17)
# cnn-toys
Convolutional neural networks are fairly simple, and it's easy to apply them. However, it's also easy to lose track of which architectures work well for different applications.
I want to use this repo to play with different applications of CNNs. That way, I gain some intuition for when to use transposed convolutions, upsampling, residual connections, pixel CNNs, leaky ReLUs, etc.
# Contents
* [colorize](https://github.com/unixpickle/cnn-toys/blob/master/cnn_toys/colorize) - Grayscale -> color predictor. *Current status:* the model sometimes colors skies in correctly, but it's generally pretty terrible.
* [cyclegan](https://github.com/unixpickle/cnn-toys/blob/master/cnn_toys/cyclegan) - A re-implementation of [CycleGAN](https://github.com/unixpickle/cnn-toys/blob/master/https://github.com/junyanz/CycleGAN). *Current status:* works fairly well.
* [real_nvp](https://github.com/unixpickle/cnn-toys/blob/master/cnn_toys/real_nvp) - A re-implementation of [real NVP](https://github.com/unixpickle/cnn-toys/blob/master/https://arxiv.org/abs/1605.08803). *Current status:* works on problems that I've tested it on. Generally requires messing with the architecture (usually making it way deeper).
近期下载者：
相关文件：
收藏者：