refact: fix docxs (#360)

Author: ganler

README.md

@@ -20,6 +20,7 @@
  <a href="#Documentation">Documentation</a> •
  <a href="#Quick-Start-with-Docker">Quick-Start with Docker</a> •
  <a href="#Performance">Performance</a> •
+ <a href="#Accuracy">Accuracy</a> •
  <a href="#License">License</a>
 </p>
 
@@ -87,16 +88,17 @@ We compare the prediction performance of HyperPose with [OpenPose 1.6](https://g
 | OpenPifPaf (ResNet50) | 97.6 MB | 432 x 368 | **44.16 FPS** | 14.5 FPS (OpenPifPaf) |
 
 ## Accuracy
+
 We evaluate accuracy of pose estimation models developed by hyperpose (mainly over Mscoco2017 dataset). the development environment is Ubuntu16.04, with 4 V100-DGXs and 24 Intel Xeon CPU. The training procedure takes 1~2 weeks using 1 V100-DGX for each model. (If you want to train from strach, loading the pretrained backbone weight is recommended.)
 
 | HyperPose Configuration | DNN Size | Input Size | Evaluate Dataset | Accuracy-hyperpose (Iou=0.50:0.95) | Accuracy-original (Iou=0.50:0.95) |
 | -------------------- | ---------- | ------------- | ---------------- | --------------------- | ----------------------- |
-| Openpose (vgg19) | 199 MB | 432 x 368 | Mscoco2014(random 1160 images) | 57.0 map | 58.4 map |
-| LightweightOpenpose (dailated mobilenet) | 17.7 MB | 432 x 368 | Mscoco2017(all 5000 images) | 46.1 map | 42.8 map |
-| LightweightOpenpose (mobilenet-thin) | 17.4 MB | 432 x 368 | Mscoco2017(all 5000 images) | 44.2 map | 28.06 map (Mscoco2014) |
-| LightweightOpenpose (tinyvgg) | 23.6 MB | 432 x 368 | Mscoco2017(all 5000 images) | 47.3 map | - |
-| LightweightOpenpose (resnet50) | 42.7 MB | 432 x 368 | Mscoco2017(all 5000 images) | 48.2 map | - |
-| PoseProposal (resnet18) | 45.2 MB | 384 x 384 | MPII(all 2729 images) | 54.9 map (Pckh) | 72.8 map (Pckh)|
+| OpenPose (VGG19) | 199 MB | 432 x 368 | MSCOCO2014 (random 1160 images) | 57.0 map | 58.4 map |
+| LightweightOpenPose (Dilated MobileNet) | 17.7 MB | 432 x 368 | MSCOCO2017(all 5000 img.) | 46.1 map | 42.8 map |
+| LightweightOpenPose (MobileNet-Thin) | 17.4 MB | 432 x 368 | MSCOCO2017 (all 5000 img.) | 44.2 map | 28.06 map (MSCOCO2014) |
+| LightweightOpenPose (tiny VGG) | 23.6 MB | 432 x 368 | MSCOCO2017 (all 5000 img.) | 47.3 map | - |
+| LightweightOpenPose (ResNet50) | 42.7 MB | 432 x 368 | MSCOCO2017 (all 5000 img.) | 48.2 map | - |
+| PoseProposal (ResNet18) | 45.2 MB | 384 x 384 | MPII (all 2729 img.) | 54.9 map (PCKh) | 72.8 map (PCKh)|
 
 
 </a>

docs/markdown/install/prediction.md

@@ -1,9 +1,7 @@
 # C++ Prediction Library Installation
 
 Note that C++ prediction library requires NVidia GPU acceleration.
- 
-Thought it is built to be platform-independent, the C++ library is mostly tested on Linux Platforms.
-So we recommend you to build it on Linux platforms. 
+HyperPose is developed and frequently tested on Linux platforms. Hence, we recommend you to build HyperPose on Linux.
 
 ## Docker Environment Installation (RECOMMENDED)
 
@@ -40,14 +38,14 @@ docker run --rm --gpus all $(USER_DEF_NAME)
 
 ### Prerequisites
 
-* C++ 17 Compiler. (g++7, clang++4.0, MSVC19.0 or newer)
+* C++ 17 Compiler. (g++7, clang++5.0, MSVC19.0 or newer)
 * CMake 3.5+ 
 * Third-Party
  * OpenCV3.2+. (**[OpenCV 4+](https://docs.opencv.org/trunk/d7/d9f/tutorial_linux_install.html) is highly recommended**)
  * [CUDA 10.2](https://developer.nvidia.com/cuda-downloads), [CuDNN 7.6.5](https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html), [TensorRT 7.0](https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html).
  * gFlags(for command-line tool/examples/tests)
 
-> Other versions of the packages may also work but not tested.
+> Packages of other versions might also work but not tested.
 
 > **About TensorRT installation**
 >

docs/markdown/performance/prediction.md

@@ -2,21 +2,22 @@
 
 ## Result
 
-| Method | Backbone Size | Network Resolution | Operator API / FPS | Stream API / FPS | Other Framework / FPS | Batch Size |
-| ------------------------ | ------------- | ------------------ | ------------------ | ---------------- | --------------------- | ---------- |
-| OpenPose COCO | 209.3MB | 656 x 368 | 19.78 | 27.32 | 8 (OpenPose) | 8 |
-| Tiny VGG + PAF | 34.7 MB | 384 x 256 | 66.62 | 124.925 | / | 8 |
-| MobileNet + PAF | 17.9 MB | 432 x 368 | 50.89 | 84.32 | / | 8 |
-| ResNet50 + PAF | 45.0 MB | 432 x 368 | 50.89 | 84.32 | 8.5 (TF-Pose) | 8 |
-| ResNet18 + Pose Proposal | 50.3 MB | 384 x 384 | 212.42 | 349.17 | / | 64 |
+
+We compare the prediction performance of HyperPose with [OpenPose 1.6](https://github.com/CMU-Perceptual-Computing-Lab/openpose) and [TF-Pose](https://github.com/ildoonet/tf-pose-estimation). 
+We implement the OpenPose algorithms with different configurations in HyperPose. 
+
+The test-bed has Ubuntu18.04, 1070Ti GPU, Intel i7 CPU (12 logic cores). 
+
+| HyperPose Configuration | DNN Size | Input Size | HyperPose | Baseline |
+| --------------- | ------------- | ------------------ | ------------------ | --------------------- |
+| OpenPose (VGG) | 209.3MB | 656 x 368 | **27.32 FPS** | 8 FPS (OpenPose) |
+| OpenPose (TinyVGG) | 34.7 MB | 384 x 256 | **124.925 FPS** | N/A |
+| OpenPose (MobileNet) | 17.9 MB | 432 x 368 | **84.32 FPS** | 8.5 FPS (TF-Pose) |
+| OpenPose (ResNet18) | 45.0 MB | 432 x 368 | **62.52 FPS** | N/A |
+| OpenPifPaf (ResNet50) | 97.6 MB | 97 x 129 | **178.6 FPS** | 35.3 |
 
 > **Environment**: System@Ubuntu18.04, GPU@1070Ti, CPU@i7(12 logic cores). 
 >
 > **Tested Video Source**: Crazy Updown Funk(resolution@640x360, frame_count@7458, source@[YouTube](https://www.youtube.com/watch?v=2DiQUX11YaY))
 
 > OpenPose performance is not tested with batch processing as it seems not to be implemented. (see [here](https://github.com/CMU-Perceptual-Computing-Lab/openpose/issues/100)) 
-
-## Suggestions
-
-- PAF post processing is slow. Batch processing will not accelerate PAF and will bring little improvement in the speed.
-- And Pose Proposal post processing is fast(over 8k FPS in single core). So any optimization(e.g. batch processing) in DNN inference will be remarkable for the throughput of the pipeline. For example, using batch size 8 we got 164 FPS, using batch size 64 we got 349 FPS, and using batch size 128 we got 383 FPS.