IJCAl2025
/
WUJI_INF
Not watched
6
0
Fork 0
Code
Releases 0 Wiki Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
Branch: master
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'master'
${ noResults }
WUJI_INF/Track#1-Task#1.md

9.4 KiB
Raw Permalink Blame History

📌 Track 1 - Task 1: Spike-guided deblur with hybrid RGB-Spike Cameras

Track 1 of the SpikeCV Wuji Challenge includes a Spike-guided deblur task designed to combine the advantages of spike cameras and traditional cameras, enabling the generation of high-fidelity color Name of Scenes from continuous spike streams.

📁 Dataset Overview

We provide two types of spike datasets for this task:

  1. Synthetic Spike Dataset based on GoPro Dataset
  2. Synthetic Spike Dataset based on X4K1000FPS Dataset
  3. Real-Captured Spike and Blurred Image Dataset

📦 Download address:
https://openi.pcl.ac.cn/IJCAl2025/WUJI_INF/datasets

The corresponding relationship between the dataset and the file:

Dataset (Part) File
Training set of the GoPro-based dataset GOPRO_train.zip
Evaluation set of the GoPro-based dataset GOPRO_test.zip
Training set of the X4K1000FPS-based dataset train_.zip
Evaluation set of the X4K1000FPS-based dataset val_.zip
Real-captured spike and blurred image dataset RSB.zip

🔷 1. Synthetic Spike Dataset Based on GoPro Dataset

The training data is in the root folder of unzipped GOPRO_train.zip, the folder structure is

---- GOPRO_Large
		|---- train
		|       |---- [Name of Scene1]
		|	   |			|---- blur
		|	   |			|  	  |---- 000047.png
		|	   |		    |	    |---- 000048.png
		|	   |		    |	    |---- ...
		|       |			|---- blur_gamma
		|	   |			|  	  |---- 000047.png
		|	   |		    |	    |---- 000048.png
		|	   |		    |	    |---- ...
		|	   |            |----sharp
		|	   |			|  	  |---- 000047.png
		|	   |		    |	    |---- 000048.png
		|	   |		    |	    |---- ...
		|	   |---- [Name of Scene2]
		|	   |---- ...
		|---- GOPRO_Large_spike_seq
		|       |---- train
		|	   |		|---- [Name of Scene1]
		|	   |		|			|---- spike
		|	   |		|			|		  |---- 000047.dat
		|	   |		|			|		  |---- 000048.dat
		|	   |		|			|		  |---- ....
		|	   |		|---- [Name of Scene2]
		|	   |		|---- ...

A demo code for reading data during training can be found in deblur_read_gopro_train_data.py

The evaluation data is in the root folder of unzipped GOPRO_test.zip, the folder structure is

---- GOPRO_Large
		|---- test
		|       |---- [Name of Scene1]
		|	   |			|---- blur
		|	   |			|  	  |---- 000001.png
		|	   |		    |	    |---- 000002.png
		|	   |		    |	    |---- ...
		|       |			|---- blur_gamma
		|	   |			|  	  |---- 000001.png
		|	   |		    |	    |---- 000002.png
		|	   |		    |	    |---- ...
		|	   |---- [Name of Scene2]
		|	   |---- ...
		|---- GOPRO_Large_spike_seq
		|       |---- test
		|	   |		|---- [Name of Scene1]
		|	   |		|			|---- spike
		|	   |		|			|		  |---- 000001.dat
		|	   |		|			|		  |---- 000002.dat
		|	   |		|			|		  |---- ....
		|	   |		|---- [Name of Scene2]
		|	   |		|---- ...

A demo code for reading data during evaluation can be found in deblur_read_gopro_test_data.py

The dataset interface of the training / evaluation set can be found in

deblur_gopro_dataset.py (Dataset Wrapper)

deblur_gopro_sequence.py (Detailed Dataset Interface)

🔷 2. Synthetic Spike Dataset Based on X4K1000FPS Dataset

The training data is in the root folder of unzipped train_.zip, the folder structure is

---- train_blurry_33
		|---- [Name of Scene1]
		|			|---- [Sub Name of Scene1]
		|			|				|---- PNGs
		|			|---- [Sub Name of Scene2]
		|			|				|---- PNGs
		|			|...
		|----[Name of Scene2]
		|....
---- train_sharp_33
		|---- [Name of Scene1]
		|			|---- [Sub Name of Scene1]
		|			|				|---- PNGs
		|			|---- [Sub Name of Scene2]
		|			|				|---- PNGs
		|			|...
		|----[Name of Scene2]
		|....
---- train_spike_2xds
		|---- [Name of Scene1]
		|			|---- [Sub Name of Scene1]
		|			|				|---- spike.dat
		|			|---- [Sub Name of Scene2]
		|			|				|---- spike.dat
		|			|...
		|----[Name of Scene2]
		|....

A demo code for reading data during training can be found in deblur_read_x4k_spk_train_data.py

The evaluation data is in the root folder of unzipped val_.zip, the folder structure is

---- train_blurry_33
		|---- [Name of Scene1]
		|			|---- [Sub Name of Scene1]
		|			|				|---- PNGs
		|			|---- [Sub Name of Scene2]
		|			|				|---- PNGs
		|			|...
		|----[Name of Scene2]
		|....
---- train_spike_2xds
		|---- [Name of Scene1]
		|			|---- [Sub Name of Scene1]
		|			|				|---- spike.dat
		|			|---- [Sub Name of Scene2]
		|			|				|---- spike.dat
		|			|...
		|----[Name of Scene2]
		|....

A demo code for reading data during evaluation can be found in deblur_read_x4k_spk_val_data.py

The dataset interface of the training / evaluation set can be found in

deblur_x4k_dataset.py (Dataset Wrapper)

deblur_x4k_sequence.py (Detailed Dataset Interface)

🔷 3. Real-Captured Spike and Blurred Image (RSB) Dataset

The real-captured spike and blurred image dataset is in the root folder of unzipped RSB.zip, the folder structure is

ego_calibration (Scene 1)
				|---- rgb
				|		|---- 000.jpg
				|		|---- 001.jpg
				|		|---- ...
				|---- spike
				|		|---- 000.dat
				|		|---- 001.dat
				|		|---- ...

ego_doll (Scene 2)
				|---- rgb
				|		|---- 000.jpg
				|		|---- 001.jpg
				|		|---- ...
				|---- spike
				|		|---- 000.dat
				|		|---- 001.dat
				|		|---- ...
...

kid (Scene 5)

Different scenarios have different numbers of blurred image-spike dat-file pairs:

Index Name of the Scene Number of image-spike pairs Index of deblurred images that need to be submitted
1 ego_calibration 62 005, 010, 015, 020, ..., 055, 060
2 ego_doll 63 005, 010, 015, 020, ..., 055, 060
3 ego_earphone 62 005, 010, 015, 020, ..., 055, 060
4 face 1 001
5 kid 1 001

Note that for a dat file with the same index as the blurred image, the 151st frame, i.e., the frame with the serial number 150 in Python, corresponds to the moment of the blurred image.

📂 Submission Format

A zip file should be submitted, the name of the zip file should be the team name. The structure is:

---- GoPro
		|---- [Name of Scene1]
		|			|---- 000001.dat
		|			|---- ...
		|			|---- 000100.dat
		|---- [Name of Scene2]
		|			|---- 000001.dat
		|			|---- ...
		|			|---- 000100.dat
		|...
		|---- [Name of Scene 11]

---- X4K1000FPS
		|---- Type1
		|		 |---- [Name of Scene 1]
		|		 |				|---- [Name of Image 1]
		|		 |				|...
		|		 |				|---- [Name of Image N]
		|		 |...
		|		 |---- [Name of Scene 5]
		|---- Type2
		|		 |---- [Name of Scene 1]
		|		 |				|---- [Name of Image 1]
		|		 |				|...
		|		 |				|---- [Name of Image N]
		|		 |...
		|		 |---- [Name of Scene 5]
		|---- Type3
		|		 |---- [Name of Scene 1]
		|		 |				|---- [Name of Image 1]
		|		 |				|...
		|		 |				|---- [Name of Image N]
		|		 |...
		|		 |---- [Name of Scene 5]

---- RSB
		|---- ego_calibration
		|			|---- 005.png
		|			|---- 010.png
		|			|...
		|			|---- 060.png
		|---- ego_doll
		|			|---- 005.png
		|			|---- 010.png
		|			|...
		|			|---- 060.png
		|---- ego_earphone
		|			|---- 005.png
		|			|---- 010.png
		|			|...
		|			|---- 060.png
		|----     face
		|			|---- 001.png
		|---- 	   kid
		|			|---- 001.png

Compress and upload the results.zip as your final result.

Note that in this track, the spatial size of the deblurred rgb images should be the same with the shape of the blurred rgb images.

🛠 Reference Code

The evaluation metrics are same with the Task4: Color Reconstruction

You can use the following official scripts to help with the metrics for evaluation:

🙏 Acknowledgements

We would like to thank Shiyan Chen and Kang Chen for his support and for providing access to the high-quality datasets used in this challenge.

📚 Reference

Shiyan Chen et al.
"Enhancing Motion Deblurring in High-Speed Scenes with Spike Streams"
NeurIPS 2023

Please read the following content carefully:

Dear OpenI User

Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.

For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》

Agree and continue
Disagree, exit