Immortal tracker

Last update: Dec 03, 2022

Related tags

Overview

Immortal_tracker

Prerequisite

Our code is tested for Python 3.6.
To install required liabraries:

pip install -r requirements.txt

Waymo Open Dataset

Prepare dataset & off-the-shelf detections

Download WOD perception dataset:

#Waymo Dataset         
└── waymo
       ├── training (not required)  
       ├── validation   
       ├── testing

To extract timestamp infos/ego infos from .tfrecord files, run the following:

bash preparedata/waymo/waymo_preparedata.sh  /
   
    /waymo

Run the following to convert detection results into to .npz files. The detection results should be in official WOD submission format(.bin)
We recommand you to use CenterPoint(two-frame model for tracking) detection results for reproducing our results. Please follow https://github.com/tianweiy/CenterPoint or email its author for CenterPoint detection results.

bash preparedata/waymo/waymo_convert_detection.sh 
   
    /detection_result.bin cp

#you can also use other detections:
#bash preparedata/waymo/waymo_convert_detection.sh

Inference

Use the following command to start inferencing on WOD. The validation set is used by default.

python main_waymo.py --name immortal --det_name cp --config_path configs/waymo_configs/immortal.yaml --process 8

Evaluation with WOD official devkit:

Follow https://github.com/waymo-research/waymo-open-dataset to build the evaluation tools and run the following command for evaluation:

#Convert the tracking results into .bin file
python evaluation/waymo/pred_bin.py --name immortal
#For evaluation

   
    /bazel-bin/waymo_open_dataset/metrics/tools/compute_tracking_metrics_main mot_results/waymo/validation/immortal/bin/pred.bin 
    
     /validation_gt.bin

nuScenes Dataset

Prepare dataset & off-the-shelf detections

Download nuScenes perception dataset

# For nuScenes Dataset         
└── NUSCENES_DATASET_ROOT
       ├── samples       
       ├── sweeps       
       ├── maps         
       ├── v1.0-trainval 
       ├── v1.0-test

To extract timestamp infos/ego infos, run the following:

bash preparedata/nuscenes/nu_preparedata.sh 
   
    /nuscenes

Run the following to convert detection results into to .npz files. The detection results should be in official nuScenes submission format(.json)
We recommand you to use centerpoint(two-frame model for tracking) detection results for reproducing our results.

bash preparedata/nuscenes/nu_convert_detection.sh  
   
    /detection_result.json cp

#you can also use other detections:
#bash preparedata/nuscenes/nu_convert_detection.sh

Inference

Use the following command to start inferencing on nuScenes. The validation set is used by default.

python main_nuscenes.py --name immortal --det_name cp --config_path configs/nu_configs/immortal.yaml --process 8

Evaluation with nuScenes official devkit:

Follow https://github.com/nutonomy/nuscenes-devkit to build the official evaluation tools for nuScenes. Run the following command for evaluation:

/nuscenes ">

#To convert tracking results into .json format
bash evaluation/nuscenes/pipeline.sh immortal
#To evaluate
python 
   
    /nuscenes-devkit/python-sdk/nuscenes/eval/tracking/evaluate.py \
"./mot_results/nuscenes/validation_2hz/immortal/results/results.json" \
--output_dir "./mot_results/nuscenes/validation_2hz/immortal/results" \
--eval_set "val" \
--dataroot 
    
     /nuscenes

Immortal tracker

Related tags

Overview

Immortal_tracker

Prerequisite

Waymo Open Dataset

Prepare dataset & off-the-shelf detections

Download WOD perception dataset:

Inference

Evaluation with WOD official devkit:

nuScenes Dataset

Prepare dataset & off-the-shelf detections

Download nuScenes perception dataset

Inference

Evaluation with nuScenes official devkit:

Owner

Implementation of Change-Based Exploration Transfer (C-BET)

[NeurIPS 2021] Code for Unsupervised Learning of Compositional Energy Concepts

Code samples for my book "Neural Networks and Deep Learning"

An open-source project for applying deep learning to medical scenarios

An Implementation of SiameseRPN with Feature Pyramid Networks

Learning Efficient Online 3D Bin Packing on Packing Configuration Trees

Parsing, analyzing, and comparing source code across many languages

Telegram chatbot created with deep learning model (LSTM) and telebot library.

MMFlow is an open source optical flow toolbox based on PyTorch

Code for "Learning From Multiple Experts: Self-paced Knowledge Distillation for Long-tailed Classification", ECCV 2020 Spotlight

Pytorch implementation of paper "Efficient Nearest Neighbor Language Models" (EMNLP 2021)

[NeurIPS'21] "AugMax: Adversarial Composition of Random Augmentations for Robust Training" by Haotao Wang, Chaowei Xiao, Jean Kossaifi, Zhiding Yu, Animashree Anandkumar, and Zhangyang Wang.

A python implementation of Physics-informed Spline Learning for nonlinear dynamics discovery

3ds-Ghidra-Scripts - Ghidra scripts to help with 3ds reverse engineering

Immortal tracker

Mixup for Supervision, Semi- and Self-Supervision Learning Toolbox and Benchmark

CRNN With PyTorch

Ultra-lightweight human body posture key point CNN model. ModelSize:2.3MB HUAWEI P40 NCNN benchmark: 6ms/img,

Histocartography is a framework bringing together AI and Digital Pathology

This is an example implementation of the paper "Cross Domain Robot Imitation with Invariant Representation".