3rd Place Solution of Traffic4Cast 2021 Core Challenge

This is the code for our solution to the NeurIPS 2021 Traffic4Cast Core Challenge.

Paper

Our solution is described in the "Solving Traffic4Cast Competition with U-Net and Temporal Domain Adaptation" paper.

If you wish to cite this code, please do it as follows:

@misc{konyakhin2021solving,
      title={Solving Traffic4Cast Competition with U-Net and Temporal Domain Adaptation}, 
      author={Vsevolod Konyakhin and Nina Lukashina and Aleksei Shpilman},
      year={2021},
      eprint={2111.03421},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Competition and Demonstration Track @ NeurIPS 2021

Learnt parameters

The models' learnt parameters are available by the link: https://drive.google.com/file/d/1zD0CecX4P3v5ugxaHO2CQW9oX7_D4BCa/view?usp=sharing
Please download the archive and unzip it into the weights folder of the repository, so its structure looks like the following:

├── ...
├── traffic4cast
├── weights
│   ├── densenet                 
│   │   ├── BERLIN_1008_1430_densenet_unet_mse_best_val_loss_2019=78.4303.pth                     
│   │   ├── CHICAGO_1010_1730_densenet_unet_mse_best_val_loss_2019=41.1579.pth
│   │   └── MELBOURNE_1009_1619_densenet_unet_mse_best_val_loss_2019=25.7395.pth    
│   ├── effnetb5
│   │   ├── BERLIN_1008_1430_efficientnetb5_unet_mse_best_val_loss_2019=80.3510.pth    
│   │   ├── CHICAGO_1012_1035_efficientnetb5_unet_mse_best_val_loss_2019=41.6425.pth
│   │   ├── ISTANBUL_1012_2315_efficientnetb5_unet_mse_best_val_loss_2019=55.7918.pth    
│   │   └── MELBOURNE_1010_0058_efficientnetb5_unet_mse_best_val_loss_2019=26.0132.pth    
│   └── unet
│       ├── BERLIN_0806_1425_vanilla_unet_mse_best_val_loss_2019=0.0000_v5.pth    
│       ├── CHICAGO_0805_0038_vanilla_unet_mse_best_val_loss_2019=42.6634.pth
│       ├── ISTANBUL_0805_2317_vanilla_unet_mse_best_val_loss_2019=0.0000_v4.pth
│       └── MELBOURNE_0804_1942_vanilla_unet_mse_best_val_loss_2019=26.7588.pth
├── ...

Submission reproduction

To generate the submission file, please run the following script:

# $1 - absolute path to the dataset, $2 device to run inference
sh submission.sh {absolute path to dataset} {cpu, cuda}
# Launch example
sh submission.sh /root/data/traffic4cast cuda

The above sctipt generates the submission file submission/submission_all_unets_da_none_mpcpm1_mean_temporal_{date}.zip, which gave us the best MSE of 49.379068541527 on the final leaderboard.

3rd Place Solution of the Traffic4Cast Core Challenge @ NeurIPS 2021

Related tags

Overview

3rd Place Solution of Traffic4Cast 2021 Core Challenge

Paper

Learnt parameters

Submission reproduction

Owner

Lightweight mmm - Lightweight (Bayesian) Media Mix Model

Official implementation of "One-Shot Voice Conversion with Weight Adaptive Instance Normalization".

Hierarchical probabilistic 3D U-Net, with attention mechanisms (—𝘈𝘵𝘵𝘦𝘯𝘵𝘪𝘰𝘯 𝘜-𝘕𝘦𝘵, 𝘚𝘌𝘙𝘦𝘴𝘕𝘦𝘵) and a nested decoder structure with deep supervision (—𝘜𝘕𝘦𝘵++).

Using Machine Learning to Create High-Res Fine Art

Example repository for custom C++/CUDA operators for TorchScript

NeuralDiff: Segmenting 3D objects that move in egocentric videos

Betafold - AlphaFold with tunings

This repo contains the implementation of the algorithm proposed in Off-Belief Learning, ICML 2021.

You Only Look One-level Feature (YOLOF), CVPR2021, Detectron2

Code repository for "Reducing Underflow in Mixed Precision Training by Gradient Scaling" presented at IJCAI '20

Implementation of "GNNAutoScale: Scalable and Expressive Graph Neural Networks via Historical Embeddings" in PyTorch

The implementation of PEMP in paper "Prior-Enhanced Few-Shot Segmentation with Meta-Prototypes"

Plotting points that lie on the intersection of the given curves using gradient descent.

SVG Icon processing tool for C++

Loopy belief propagation for factor graphs on discrete variables, in JAX!

Checkout some cool self-projects you can try your hands on to curb your boredom this December!

Read number plates with https://platerecognizer.com/

Reverse engineering recurrent neural networks with Jacobian switching linear dynamical systems

Evidential Softmax for Sparse Multimodal Distributions in Deep Generative Models

PointNetVLAD: Deep Point Cloud Based Retrieval for Large-Scale Place Recognition, CVPR 2018