Model zoo for classification (dmlc#53)

* improve imagenet scripts

* add history plot utils

* add history plot in cifar script

* fix plot on server (no X)

* half finished tutorial, needs pretrained model first

* add best model saving for imagenet

* fix bug

* fix

* fix

* add demo

* fix

* add pretrained model accuracy

* add pretrained resnet for cifar10

* add wrn pretrained-model

* add pretrained model to wrn, modify demo

* reorg scripts/classification

* rename files, update doc

* improve

* add doc

* change to rst

* improve plot function

* merge

* move doc to script

* reorg tutorial structure

* restruct sphinx

* add finetune

* local changes

* improve

* fix

* fix

* fix plot

* fix

* fix demo

* rename

* add prepare data

* fix finetune

* change to minc data

* fix augmentation

* finish finetune minc 2500

* improve docs

* add gitignore for docs/

* add finetune

* rollback gitignore

* fix

* switch py code + cooments

* fix

* fix script

* add cifar demo and finetune as py

* mask transfer learning

* improve doc

* improve

* improve

* add runnable finetune tutorial

* add comments to finetune tutorial

* fix

* fix

* add imagenet training script, fix api doc

* improve tutorial

* improve

* improve

* refine cifar10 demo

* add imagenet demo

* improve cifar10 training

* improve transfer learning

* improve imagenet training

* refine history plotting

* fix

* fix lint

* add path to plot

* fix

* improve

* improve script readme

* ordering title, improve model zoo table

* improve table

* re-define example sort key

* fix

* test table

* fix table

* narrower table

* narrow

* test adding link

* fix large table

* narrow

* add hint

* fix

* refine hint

* fix download link

docs/model_zoo/index.rst

@@ -48,12 +48,13 @@ The following table summarizes the available models and their performances in ad
 
 .. hint::
 
-    Our pre-trained models match the accuracy from "Mix-Up".
+    Our pre-trained models match the accuracy from "Mix-Up" [4]_ .
+    Please check the reference paper for further information.
 
-    The training commands assume you are under corresponding working directory:
+    The training commands work with the following scripts:
 
-    - For ``CIFAR10``, the working directory is `scripts/classification/cifar <https://github.com/dmlc/gluon-vision/tree/master/scripts/classification/cifar>`_
-    - For ``ImageNet``, the working directory is `scripts/classification/imagenet <https://github.com/dmlc/gluon-vision/tree/master/scripts/classification/imagenet>`_
+    - For vanilla training: :download:`Download train_cifar10.py<../../scripts/classification/cifar/train_cifar10.py>`
+    - For mix-up training: :download:`Download train_mixup_cifar10.py<../../scripts/classification/cifar/train_mixup_cifar10.py>`
 
 +----------------------------+----------------------------+------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 | Model                      | Acc (Vanilla/Mix-Up [4]_ ) | Training Command                                                                                                                                                                                                                                                             | Training Log                                                                                                                                                                                                                                                                   |