DeFMO: Deblurring and Shape Recovery of Fast Moving Objects (CVPR 2021)

Last update: Dec 26, 2022

Overview

Evaluation, Training, Demo, and Inference of DeFMO

DeFMO: Deblurring and Shape Recovery of Fast Moving Objects (CVPR 2021)

Denys Rozumnyi, Martin R. Oswald, Vittorio Ferrari, Jiri Matas, Marc Pollefeys

Qualitative results: https://www.youtube.com/watch?v=pmAynZvaaQ4

Pre-trained models

The pre-trained DeFMO model as reported in the paper is available here: https://polybox.ethz.ch/index.php/s/M06QR8jHog9GAcF. Put them into ./saved_models sub-folder.

Inference

For generating video temporal super-resolution:

python run.py --video example/falling_pen.avi

For generating temporal super-resolution of a single frame with the given background:

python run.py --im example/im.png --bgr example/bgr.png

Evaluation

After downloading the pre-trained models and downloading the evaluation datasets, you can run

python eval_dataset.py

Synthetic dataset generation

For the dataset generation, please download:

ShapeNetCore.v2 dataset: https://www.shapenet.org/.
Textures from the DTD dataset: https://www.robots.ox.ac.uk/~vgg/data/dtd/. The exact split used in DeFMO is from the "Neural Voxel Renderer: Learning an Accurate and Controllable Rendering Tool" model and can be downloaded here: https://polybox.ethz.ch/index.php/s/9Abv3QRm0ZgPzhK.
Backgrounds for the training dataset from the VOT dataset: https://www.votchallenge.net/vot2018/dataset.html.
Backgrounds for the testing dataset from the Sports1M dataset: https://cs.stanford.edu/people/karpathy/deepvideo/.
Blender 2.79b with Python enabled.

Then, insert your paths in renderer/settings.py file. To generate the dataset, run in renderer sub-folder:

python run_render.py

Note that the full training dataset with 50 object categories, 1000 objects per category, and 24 timestamps takes up to 1 TB of storage memory. Due to this and also the ShapeNet licence, we cannot make the pre-generated dataset public - please generate it by yourself using the steps above.

Training

Set up all paths in main_settings.py and run

python train.py

Evaluation on real-world datasets

All evaluation datasets can be found at http://cmp.felk.cvut.cz/fmo/. We provide a download_datasets.sh script to download the Falling Objects, the TbD-3D, and the TbD datasets.

Reference

If you use this repository, please cite the following publication ( https://arxiv.org/abs/2012.00595 ):

@inproceedings{defmo,
  author = {Denys Rozumnyi and Martin R. Oswald and Vittorio Ferrari and Jiri Matas and Marc Pollefeys},
  title = {DeFMO: Deblurring and Shape Recovery of Fast Moving Objects},
  booktitle = {CVPR},
  address = {Nashville, Tennessee, USA},
  month = jun,
  year = {2021}
}

Comments

Question about training set

Hi, thanks for your generous sharing.

I have a question about training set generating in your work. I generated a training set following your codes. Its size is about 100GB, far less than 1TB. Is there anything wrong?

Thanks.

opened by fan-hd 11
Apply your model on custom longer video clips

Hi thank you for releasing your code,

Can your model be applied on custom videos about high speed train crossing? Video clips last from 3 to 10 seconds, my idea was to preprocess them with your code in order to keep the same frame rate and have a better video quality for later object detection. This is an example frame from original video clip:

I tried to run your code on a video about 6 seconds and the result was a longer video (about 13min) with a lower level of detail, probably I'm doing something wrong. This is an example frame from output video clip:

How can I correctly reconstruct the quality of single frames usin all the information contained in the video?

opened by fabiozappo 4
Question about comparison with Jin et al.'s work (CVPR2018)

Hi, thank you for your interesting work! I have a question about the comparison of methods in your work. When making comparisons, did you retrain Jin et al.'s model ("Learning to Extract a Video Sequence from a Single Motion-Blurred Image" from CVPR 2018), or did you just use their pre-trained checkpoints? I couldn't find the training code on their github page.

opened by zzh-tech 2
Padding in Time-Consistency Loss

Hi,

Congratulations!

I found that "padding = tuple(side // 10 for side in sh[:2]) + (0,)" for normalized cross-correlation. Does it only implement padding to the height axis, since the padding tuple will be of size (4//10, H//10, 0)?

Thanks a lot.

opened by JLiu-Edinburgh 1
run on google colab!

I'm confused! and need to run the code on google colab or more explanation about how to implement that code in vscode or something else .if it know someone please help me

opened by ganikas 3

Releases(v1.0)

v1.0(Feb 7, 2021)

The first release of the method reported in the arxiv paper.
Source code(tar.gz)
Source code(zip)

Owner

Denys Rozumnyi

PhD student at ETH Zurich.

GitHub Repository https://github.com/rozumden/DeFMO

nnDetection is a self-configuring framework for 3D (volumetric) medical object detection which can be applied to new data sets without manual intervention. It includes guides for 12 data sets that were used to develop and evaluate the performance of the proposed method.

What is nnDetection? Simultaneous localisation and categorization of objects in medical images, also referred to as medical object detection, is of hi

365 Jan 09, 2023

Tensors and Dynamic neural networks in Python with strong GPU acceleration

PyTorch is a Python package that provides two high-level features: Tensor computation (like NumPy) with strong GPU acceleration Deep neural networks b

61.4k Jan 04, 2023

CLIP: Connecting Text and Image (Learning Transferable Visual Models From Natural Language Supervision)

CLIP (Contrastive Language–Image Pre-training) Experiments (Evaluation) Model Dataset Acc (%) ViT-B/32 (Paper) CIFAR100 65.1 ViT-B/32 (Our) CIFAR100 6

52 Jan 07, 2023

Official Implement of CVPR 2021 paper “Cross-Modal Collaborative Representation Learning and a Large-Scale RGBT Benchmark for Crowd Counting”

RGBT Crowd Counting Lingbo Liu, Jiaqi Chen, Hefeng Wu, Guanbin Li, Chenglong Li, Liang Lin. "Cross-Modal Collaborative Representation Learning and a L

37 Dec 08, 2022

DeFMO: Deblurring and Shape Recovery of Fast Moving Objects (CVPR 2021)

Related tags

Overview

Evaluation, Training, Demo, and Inference of DeFMO

DeFMO: Deblurring and Shape Recovery of Fast Moving Objects (CVPR 2021)

Denys Rozumnyi, Martin R. Oswald, Vittorio Ferrari, Jiri Matas, Marc Pollefeys

Qualitative results: https://www.youtube.com/watch?v=pmAynZvaaQ4

Pre-trained models

Inference

Evaluation

Synthetic dataset generation

Training

Evaluation on real-world datasets

Reference

Comments

Question about training set

Apply your model on custom longer video clips

Question about comparison with Jin et al.'s work (CVPR2018)

Padding in Time-Consistency Loss

run on google colab!

Releases(v1.0)

v1.0(Feb 7, 2021)

Owner

Denys Rozumnyi

nnDetection is a self-configuring framework for 3D (volumetric) medical object detection which can be applied to new data sets without manual intervention. It includes guides for 12 data sets that were used to develop and evaluate the performance of the proposed method.

Tensors and Dynamic neural networks in Python with strong GPU acceleration

CLIP: Connecting Text and Image (Learning Transferable Visual Models From Natural Language Supervision)

Official Implement of CVPR 2021 paper “Cross-Modal Collaborative Representation Learning and a Large-Scale RGBT Benchmark for Crowd Counting”

Image marine sea litter prediction Shiny

Code for "Learning Graph Cellular Automata"

Multiview Neural Surface Reconstruction by Disentangling Geometry and Appearance

Deep Ensemble Learning with Jet-Like architecture

SMPL-X: A new joint 3D model of the human body, face and hands together

Python package for multiple object tracking research with focus on laboratory animals tracking.

Accelerate Neural Net Training by Progressively Freezing Layers

A higher performance pytorch implementation of DeepLab V3 Plus(DeepLab v3+)

Using Clinical Drug Representations for Improving Mortality and Length of Stay Predictions

Deep Networks with Recurrent Layer Aggregation

[CVPR 2021] Involution: Inverting the Inherence of Convolution for Visual Recognition, a brand new neural operator

A small library for creating and manipulating custom JAX Pytree classes

Adversarial Graph Representation Adaptation for Cross-Domain Facial Expression Recognition (AGRA, ACM 2020, Oral)

Boosted CVaR Classification (NeurIPS 2021)

MAUS: A Dataset for Mental Workload Assessment Using Wearable Sensor - Baseline system

Traditional deepdream with VQGAN+CLIP and optical flow. Ready to use in Google Colab