typos & clean-ups (including #358)

examples/A3C-Gym/train-atari.py

@@ -51,7 +51,6 @@ PREDICT_BATCH_SIZE = 15     # batch for efficient forward
 SIMULATOR_PROC = 50
 PREDICTOR_THREAD_PER_GPU = 3
 PREDICTOR_THREAD = None
-EVALUATE_PROC = min(multiprocessing.cpu_count() // 2, 20)
 
 NUM_ACTIONS = None
 ENV_NAME = None

examples/GAN/BEGAN.py

@@ -18,7 +18,8 @@ from GAN import GANModelDesc, GANTrainer, MultiGPUGANTrainer
 Boundary Equilibrium GAN.
 See the docstring in DCGAN.py for usage.
 
-A pretrained model on CelebA is at https://drive.google.com/open?id=0B5uDfUQ1JTglUmgyZV8zQmNOTVU
+A pretrained model on CelebA is at
+https://drive.google.com/open?id=0B5uDfUQ1JTglUmgyZV8zQmNOTVU
 """
 
 

examples/GAN/CycleGAN.py

@@ -20,7 +20,10 @@ from GAN import GANTrainer, GANModelDesc
 """
 1. Download the dataset following the original project: https://github.com/junyanz/CycleGAN#train
 2. ./CycleGAN.py --data /path/to/datasets/horse2zebra
-Training and testing visuliazations will be in tensorboard.
+Training and testing visualizations will be in tensorboard.
+
+This implementation doesn't use fake sample buffer.
+It's not hard to add but I didn't observe any difference with it.
 """
 
 SHAPE = 256

examples/GAN/README.md

@@ -37,7 +37,7 @@ Reproduce DCGAN following the setup in [dcgan.torch](https://github.com/soumith/
 
 ## Image2Image.py
 
-Image-to-Image following the setup in [pix2pix](https://github.com/phillipi/pix2pix).
+Image-to-Image translation following the setup in [pix2pix](https://github.com/phillipi/pix2pix).
 
 For example, with the cityscapes dataset, it learns to generate semantic segmentation map of urban scene:
 
@@ -71,5 +71,6 @@ Some BEGAN samples:
 ## CycleGAN.py, DiscoGAN-CelebA.py
 
 Reproduce CycleGAN with the original datasets, and DiscoGAN on CelebA. They are pretty much the same idea with different architecture.
+CycleGAN horse-to-zebra in tensorboard:
 
 ![cyclegan-sample](demo/CycleGAN-horse2zebra.jpg)

examples/HED/README.md

@@ -27,7 +27,7 @@ To start training:
 ```bash
 ./hed.py --load vgg16.npy
 ```
-It takes about 100k steps (~10 hour on a TitanX) to reach a reasonable performance.
+It takes about 100k steps (~10 hours on a TitanX) to reach a reasonable performance.
 
 To inference (produce a heatmap at each level at out*.png):
 ```bash
@@ -41,3 +41,4 @@ cat train_log/hed/stat.json | jq '.[] |
 "\(.xentropy1)\t\(.xentropy2)\t\(.xentropy3)\t\(.xentropy4)\t\(.xentropy5)\t\(.xentropy6)"' -r | \
 				tpk-plot-point --legend 1,2,3,4,5,final --decay 0.8
 ```
+Or just open tensorboard.