The official implementation of NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021]. https://arxiv.org/pdf/2101.12378.pdf

Angtian Wang

Last update: Nov 23, 2022

Related tags

Overview

NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021]

Release Notes

The offical PyTorch implementation of NeMo, published on ICLR 2021. NeMo achieves robust 3D pose estimation method by performing render-and-compare on the level of neural network features. The figure shows a dynamic example of the pose optimization process of NeMo. Top-left: the input image; Top-right: A mesh superimposed on the input image in the predicted 3D pose. Bottom-left: The occluder location as predicted by NeMo, where yellow is background, green is the non-occluded area and red is the occluded area of the object. Bottom-right: The loss landscape as a function of each camera parameter respectively. The colored vertical lines demonstrate the current prediction and the ground-truth parameter is at center of x-axis.

Installation

The code is tested with python 3.7, PyTorch 1.5 and PyTorch3D 0.2.0.

Clone the project and install requirements

git clone https://github.com/Angtian/NeMo.git
cd NeMo
pip install -r requirements.txt

Running NeMo

We provide the scripts to train NeMo and to perform inference with NeMo on Pascal3D+ and the Occluded Pascal3D+ datasets. For more details about the OccludedPascal3D+ please refer to this Github repo: OccludedPASCAL3D.

Step 1: Prepare Datasets
Set ENABLE_OCCLUDED to "true" if you need evaluate NeMo under partial occlusions. You can change the path to the datasets in the file PrepareData.sh, after downloading the data. Otherwise this script will automatically download datasets.
Then run the following commands:

chmod +x PrepareData.sh
./PrepareData.sh

Step 2: Training NeMo
Modify the settings in TrainNeMo.sh.
GPUS: set avaliable GPUs for training depending on your machine. The standard setting uses 7 gpus (6 for the backbone, 1 for the feature bank). If you have only 4 GPUs available, we suggest to turn off the "--sperate_bank" in training stage.
MESH_DIMENSIONS: "single" or "multi".
TOTAL_EPOCHS: The default setting is 800 epochs, which takes 3 to 4 days to train on an 8 GPUs machine. However, 400 training epochs could already yield good accuracy. The final performance for the raw Pascal3D+ over train epochs (SingleCuboid):

Training Epochs	200	400	600	800
Acc Pi / 6	82.4	84.4	84.8	85.5
Acc Pi / 18	57.1	59.2	59.6	60.2

Then, run these commands:

chmod +x TrainNeMo.sh
./TrainNeMo.sh

Step 2 (Alternative): Download Pretrained Model
Here we provide the pretrained NeMo Model and backbone for the "SingleCuboid" setting. Run the following commands to download the pretrained model:

wget --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate 'https://docs.google.com/uc?export=download&id=1X1NCx22TFGJs108TqDgaPqrrKlExZGP-' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=1X1NCx22TFGJs108TqDgaPqrrKlExZGP-" -O NeMo_Single_799.zip
unzip NeMo_Single_799.zip

Step 3: Inference with NeMo
The inference stage includes feature extraction and pose optimization. The pose optimization conducts render-and-compare on the neural features w.r.t. the camera pose iteratively. This takes some time to run on the full dataset (3-4 hours for each occlusion level on a 8 GPU machine).
To run the inference, you need to first change the settings in InferenceNeMo.sh:
MESH_DIMENSIONS: Set to be same as the training stage.
GPUS: Our implemention could either utilize 4 or 8 GPUs for the pose optimization. We will automatically distribute workloads over available GPUs and run the optimization in parallel.
LOAD_FILE_NAME: Change this setting if you do not train 800 epochs, e.g. train NeMo for 400 -> "saved_model_%s_399.pth".

Then, run these commands to conduct NeMo inference on unoccluded Pascal3D+:

chmod +x InferenceNeMo.sh
./InferenceNeMo.sh

To conduct inference on the occluded-Pascal3D+ (Note you need enable to create OccludedPascal3D+ dataset during data preparation):

./InferenceNeMo.sh FGL1_BGL1
./InferenceNeMo.sh FGL2_BGL2
./InferenceNeMo.sh FGL3_BGL3

Citation

Please cite the following paper if you find this the code useful for your research/projects.

@inproceedings{wang2020NeMo,
title = {NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation},
author = {Angtian, Wang and Kortylewski, Adam and Yuille, Alan},
booktitle = {Proceedings International Conference on Learning Representations (ICLR)},
year = {2021},
}

Comments

Training on KITTI3D

Hi Angtian,

As the title suggests, I was just wondering whether you think it's possible and whether you have tried to train NeMo on KITTI3D considering that the samples contain ground truth only for azimuth and not for elevation or theta.

opened by gkouros 9
Split of L0-Occluded Pascal3D+

Hi,

Thanks for the great work and sharing the code!

I've got one question about the non-occluded split of Pascal3D+. In your paper, StarMap gets 89% Acc30 while the author of StarMap only reported an accuracy at 82% in their paper, I suppose this difference is resulted from a different dataset split for evaluation. Could you tell me how you get the non-occluded test images on Pascal3D+ and is there any difference between the images in occlusion L0, L1, L2, and L3?

opened by YoungXIAO13 2

"Include method supposed to be point or bbox" error running PrepareData.sh with BboxTools 1.0.2

https://github.com/Angtian/NeMo/blob/11fe2101d35ab9eac2a2f813f33d95d616dc5d02/code/lib/MeshMemoryMap.py#L60

BBox2D.include in BboxTools 1.0.2 seems to only accept single points or a Bbox2D object (code pasted below). points_2d is a numpy array of points, which causes an error at the line above, so generate3Dpascal3D.py is failing to generate any annotations.

def include(self, other):
        """
        Check if other is inside this box. Notice include means strictly include, other could not place at the boundary
        of this bbox.
        :param other: (Bbox2D or tuple of int) bbox or point
        :return: (bool) True or False
        """
        if type(other) == Bbox2D:
            out = True
            for i in range(2):
                if self.bbox[i][0] > other.bbox[i][0]:
                    out = False
                if self.bbox[i][1] < other.bbox[i][1]:
                    out = False
            return out

        if type(other) == tuple and len(other) == 2:
            if other[0] < self.bbox[0][0] or other[0] >= self.bbox[0][1]:
                return False
            if other[1] < self.bbox[1][0] or other[1] >= self.bbox[1][1]:
                return False
            return True
        raise Exception('Include method suppose to be point or bbox, but got %s' % str(other))

opened by tomforge 1

Official Implementation for "ReStyle: A Residual-Based StyleGAN Encoder via Iterative Refinement" https://arxiv.org/abs/2104.02699

ReStyle: A Residual-Based StyleGAN Encoder via Iterative Refinement Recently, the power of unconditional image synthesis has significantly advanced th

967 Jan 4, 2023

Non-Official Pytorch implementation of "Face Identity Disentanglement via Latent Space Mapping" https://arxiv.org/abs/2005.07728 Using StyleGAN2 instead of StyleGAN

Face Identity Disentanglement via Latent Space Mapping - Implement in pytorch with StyleGAN 2 Description Pytorch implementation of the paper Face Ide

58 Dec 24, 2022

A PyTorch implementation of EventProp [https://arxiv.org/abs/2009.08378], a method to train Spiking Neural Networks

Spiking Neural Network training with EventProp This is an unofficial PyTorch implemenation of EventProp, a method to compute exact gradients for Spiki

35 Jul 29, 2022

Source code for models described in the paper "AudioCLIP: Extending CLIP to Image, Text and Audio" (https://arxiv.org/abs/2106.13043)

AudioCLIP Extending CLIP to Image, Text and Audio This repository contains implementation of the models described in the paper arXiv:2106.13043. This

458 Jan 2, 2023

source code for https://arxiv.org/abs/2005.11248 "Accelerating Antimicrobial Discovery with Controllable Deep Generative Models and Molecular Dynamics"

Accelerating Antimicrobial Discovery with Controllable Deep Generative Models and Molecular Dynamics This work will be published in Nature Biomedical

71 Nov 15, 2022

Supplementary code for the paper "Meta-Solver for Neural Ordinary Differential Equations" https://arxiv.org/abs/2103.08561

Meta-Solver for Neural Ordinary Differential Equations Towards robust neural ODEs using parametrized solvers. Main idea Each Runge-Kutta (RK) solver w

25 Aug 12, 2021

Pytorch implementation of Each Part Matters: Local Patterns Facilitate Cross-view Geo-localization https://arxiv.org/abs/2008.11646

[TCSVT] Each Part Matters: Local Patterns Facilitate Cross-view Geo-localization LPN [Paper] NEWs Prerequisites Python 3.6 GPU Memory = 8G Numpy 1.

46 Dec 14, 2022

Minimal implementation of PAWS (https://arxiv.org/abs/2104.13963) in TensorFlow.

PAWS-TF 🐾 Implementation of Semi-Supervised Learning of Visual Features by Non-Parametrically Predicting View Assignments with Support Samples (PAWS)

43 Jan 8, 2023

Unofficial implementation of "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" (https://arxiv.org/abs/2103.14030)

Swin-Transformer-Tensorflow A direct translation of the official PyTorch implementation of "Swin Transformer: Hierarchical Vision Transformer using Sh

52 Dec 29, 2022

The official implementation of NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021]. https://arxiv.org/pdf/2101.12378.pdf

Related tags

Overview

NeMo: Neural Mesh Models of Contrastive Features for Robust 3D Pose Estimation [ICLR-2021]

Release Notes

Installation

Clone the project and install requirements

Running NeMo

Citation

You might also like...

Official Implementation for "ReStyle: A Residual-Based StyleGAN Encoder via Iterative Refinement" https://arxiv.org/abs/2104.02699

Non-Official Pytorch implementation of "Face Identity Disentanglement via Latent Space Mapping" https://arxiv.org/abs/2005.07728 Using StyleGAN2 instead of StyleGAN

A PyTorch implementation of EventProp [https://arxiv.org/abs/2009.08378], a method to train Spiking Neural Networks

Source code for models described in the paper "AudioCLIP: Extending CLIP to Image, Text and Audio" (https://arxiv.org/abs/2106.13043)

source code for https://arxiv.org/abs/2005.11248 "Accelerating Antimicrobial Discovery with Controllable Deep Generative Models and Molecular Dynamics"

Supplementary code for the paper "Meta-Solver for Neural Ordinary Differential Equations" https://arxiv.org/abs/2103.08561

Pytorch implementation of Each Part Matters: Local Patterns Facilitate Cross-view Geo-localization https://arxiv.org/abs/2008.11646

Minimal implementation of PAWS (https://arxiv.org/abs/2104.13963) in TensorFlow.

Unofficial implementation of "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows" (https://arxiv.org/abs/2103.14030)

Comments

Training on KITTI3D

Split of L0-Occluded Pascal3D+

"Include method supposed to be point or bbox" error running PrepareData.sh with BboxTools 1.0.2

Owner

Angtian Wang

This is an official implementation of our CVPR 2021 paper "Bottom-Up Human Pose Estimation Via Disentangled Keypoint Regression" (https://arxiv.org/abs/2104.02300)

This repo provides the official code for TransBTS: Multimodal Brain Tumor Segmentation Using Transformer (https://arxiv.org/pdf/2103.04430.pdf).

Source Code for DialogBERT: Discourse-Aware Response Generation via Learning to Recover and Rank Utterances (https://arxiv.org/pdf/2012.01775.pdf)

PGPortfolio: Policy Gradient Portfolio, the source code of "A Deep Reinforcement Learning Framework for the Financial Portfolio Management Problem"(https://arxiv.org/pdf/1706.10059.pdf).

We have implemented shaDow-GNN as a general and powerful pipeline for graph representation learning. For more details, please find our paper titled Deep Graph Neural Networks with Shallow Subgraph Samplers, available on arXiv (https//arxiv.org/abs/2012.01380).

[PyTorch] Official implementation of CVPR2021 paper "PointDSC: Robust Point Cloud Registration using Deep Spatial Consistency". https://arxiv.org/abs/2103.05465

Re-implementation of the Noise Contrastive Estimation algorithm for pyTorch, following "Noise-contrastive estimation: A new estimation principle for unnormalized statistical models." (Gutmann and Hyvarinen, AISTATS 2010)

Code for the paper: Learning Adversarially Robust Representations via Worst-Case Mutual Information Maximization (https://arxiv.org/abs/2002.11798)

Official repository with code and data accompanying the NAACL 2021 paper "Hurdles to Progress in Long-form Question Answering" (https://arxiv.org/abs/2103.06332).

Official implementation of the paper Image Generators with Conditionally-Independent Pixel Synthesis https://arxiv.org/abs/2011.13775