* fix padding bug and add resize_short_size args

* fix typo

* pass args:settings in eval and infer

* Fix SSD file path.

* Move object_detection to ssd.

* add landmark models

* modify single queue to a queue per process for data processing

* add transforms.py and rename torchvision_reader.py to reader.py

* add datasets.py

* remove pytorch apis

* fix some small bugs

* remove torch and torchvision from requirements.txt

* modify core.CUDAPlace to fluid.CUDAPlace

* bug fix: data_dir has no effect

* bug fix: data_dir has no effect

* bug fix: modify  in directory

* bug fix: fix a typo in README.md

* update image classification pretrained models

* update softmax

* Add mixup and distill and fix bugs

* recover test pyreader

* add error tip

* update README_CN

* fix distill bug

* add reader_cv2.py

* update run.sh

* fix docs

* fix typo

* polish train.py

* fix typo

* remove commented code

* fix typo

* fix_typo

* recover scale_loss

* fix typo

* fix bug

* fix test_program bug

* add yolov3 label_smooth faq

* refine en

* fix bug in cycle and pix

* add yolov3 save_inference_model example

* add save_inference_model in infer.py

* refine link

* refine yolov3 doc en.

* fix format in README.md

* refine FAQ

* refine en doc

* add zh doc

* refine zh doc

* use zh as default

* add infer config

* add data descp

* add transforms.py and rename torchvision_reader.py to reader.py

* add datasets.py

* remove pytorch apis

* fix some small bugs

* remove torch and torchvision from requirements.txt

* modify core.CUDAPlace to fluid.CUDAPlace

fix bug of gan in python3

* add warning for class_num mismatch

* change to value error

* modify .run_ce.sh

* modify .run_ce.sh

* test=develop fix error while loading cv2.VideoCapture

* test=develop
fix bug of index out of range in SampleFrames

* fix bug

* add pix2pix to gan library

* add tsm ce

* add tsm ce

* add enable_ce in configs

* remove enable_ce in val config

this is for issue #2232
add pretrain description in readme to show that some models need pertained models to begin training.

Because WINDOWS does not support NCCL. It will raise an error when users try to run the multi-GPUs program on WINDOWS machine. So add a note to remind users to replace parallel executor to executor.

* fix video train param name epoch.

* fix for script.

* fix bugs and add python3 supports
* minor changes
* Fix misleading description in README