The repo is all about implementing hprel paper.
The current state-of-the-art is focused only on RGB and RGB-D approaches for predicting the 3D human pose. However, not using precise LiDAR depth information limits the performance and leads to very inaccurate absolute pose estimation. With LiDAR sensors becoming more affordable and common on robots and autonomous vehicle setups, we propose an end-to-end architecture using RGB and LiDAR to predict the absolute 3D human pose with unprecedented precision.
The Hperl paper has two architecture
AVOD(https://arxiv.org/pdf/1712.02294.pdf) followed by LCRNET++ (https://arxiv.org/pdf/1803.00455.pdf).
Avod is used for predicting the pedestrian in the real time followed by the LCRNET++ which derives the poses from the detected pedestrian with 13 keypoints
In this repo used dope (https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123710375.pdf) which make use of LCRNET++ as a part.
Getting Started
First the AVOD (Aggregate View of Object Detection) Architecture is cloned from the git
git clone https://github.com/kujason/avod.
Here we are mainly concern about pedestrian.We have trained by the various epoch So by taking the pedestrian config file the data is trained.
AVOD is trained with AVOD_People_example config for 70 epoch and time taken is 10 Hours
After finished of the training phase the testing phase is started.
For testing it taken 16 Hours for 110 epoch
EASY(%) MODERATE(%) HARD(%) VIEW MODE
60.80 52.81 40.88 AP - 3D
58.75 51.05 47.54 AP -BEV
Once we done the pedestrian detection move to the LCRNET++ part which contains pose estimation from detected the pedestrian.
LCRNET++(Localization, Classfication & Regression )
Architecture is cloned from the git
git clone https://github.com/naver/dope
post-processing with a modified version of LCR-Net++
Our post-processing relies on a modified version of the pose proposals integration proposed in the LCR-Net++ code. To get this code, once in the DOPE folder, please clone our modified LCR-Net++ repository:
git clone https://github.com/naver/lcrnet-v2-improved-ppi.git
To use our code on an image, use the following command:
python dope.py --model <modelname> --image <imagename>
For ex:
python dope.py --model DOPErealtime_v1_0_0 --image post_estimation.jpg
Note that DOPE predicts 3D poses for each body part in their relative coordinate systems, eg, centered on body center for bodies. For better visualization, we approximate 3D scene coordinates by finding the offsets that minimize the reprojection error based on a scaled orthographic projection model.
Final Output:
