Fix misleading description in README (#2259)

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

PaddleCV/human_pose_estimation/README.md

@@ -10,9 +10,12 @@ This is a simple demonstration of re-implementation in [PaddlePaddle.Fluid](http
 ## Requirements
 
   - Python == 2.7 or 3.6
-  - PaddlePaddle >= 1.1.0 (<= 1.3.0)
+  - PaddlePaddle >= 1.1.0 
   - opencv-python >= 3.3
 
+### Notes:
+We found that there are some issues may result in misconvergence with PaddlePaddle 1.3.0 and cuDNN-7.0. So it is recommended to use the latest version of PaddlePaddle (>= 1.4).
+
 ## Environment
 
 The code is developed and tested under 4 Tesla K40/P40 GPUS cards on CentOS with installed CUDA-9.0/8.0 and cuDNN-7.0.
@@ -91,7 +94,7 @@ python val.py --dataset 'mpii' --checkpoint 'checkpoints/pose-resnet50-mpii-384x
 ### Perform Training
 
 ```bash
-python train.py --dataset 'mpii' --data_root 'data/mpii'
+python train.py --dataset 'mpii'
 ```
 
 **Note**: Configurations for training are aggregated in the `lib/mpii_reader.py` and `lib/coco_reader.py`.

PaddleCV/human_pose_estimation/README_cn.md

@@ -12,9 +12,13 @@
 本目录下的代码均在4卡Tesla K40/P40 GPU，CentOS系统，CUDA-9.0/8.0，cuDNN-7.0环境下测试运行无误
 
   - Python == 2.7 / 3.6
-  - PaddlePaddle >= 1.1.0 (<= 1.3.0)
+  - PaddlePaddle >= 1.1.0
   - opencv-python >= 3.3
 
+### 说明
+
+目前已发现在PaddlePaddle 1.3.0 / cuDNN-7.0环境下，存在问题会导致模型训练loss不收敛。推荐使用最新版本PaddlePaddle (>= 1.4).
+
 ## MPII Val结果
 | Arch | Head | Shoulder | Elbow | Wrist | Hip | Knee | Ankle | Mean | Mean@0.1| Models |
 | ---- |:----:|:--------:|:-----:|:-----:|:---:|:----:|:-----:|:----:|:-------:|:------:|
@@ -106,4 +110,4 @@ python test.py --checkpoint 'checkpoints/pose-resnet-50-384x384-mpii'
 
 ## 引用
 
-- Simple Baselines for Human Pose Estimation and Tracking in PyTorch [`code`](https://github.com/Microsoft/human-pose-estimation.pytorch#data-preparation)
\ No newline at end of file
+- Simple Baselines for Human Pose Estimation and Tracking in PyTorch [`code`](https://github.com/Microsoft/human-pose-estimation.pytorch#data-preparation)