Peyman Milanfar

MDSP Super-Resolution And Demosaicing Datasets

The following data sets have been gathered through the past several years in the Multi-Dimensional Signal Processing Research Group (MDSP), and have been used with permission in several of our publications.

We will periodically update this page with additional new datasets as they become available for public distribution. Please be sure to include a link to this page (http://www.soe.ucsc.edu/~milanfar/software/sr-datasets.html) whenever you use any of these data in a publication.

You can also explore the MDSP Resolution Enhancement Software, with example results. For comparison and reference purposes, you may be interested in the before/after image pairs generated using our methods, which have appeared in various publications cited below.

For simultaneous demosaicing/superres see the "Raw CFA Demosaicing Data Sets" tab. The frames are available in AVI format and in MATLAB ".mat" format. Click on the "Related Paper" links for more information and the published resolution enhaced results using the corresponding data sets.

Grayscale datasets Full color datasets Raw CFA Demosaicing datasets

Grayscale Datasets

  • Text
  • Infrared "Tank" Sequence
  • Car
  • Disk
  • Alpaca (Small)
  • Alpaca
  • Emily (Small)
  • Emily
  • Face 1
  • Face 2
  • Surveillance (Rotated)
  • EIA

Text

Description:
30 uncompressed grayscale frames of size 57 x 49
Camera brand:
Olympus C-4000
Motion note:
These frames approximately follow the global translational motion model.
Credit:
Sina Farsiu

Infrared "Tank" Sequence


This dataset is unfortunately not available until we have permission to share it from the US Air Force.

Description:
16 uncompressed grayscale frames of size 64 x 64
Camera brand:
Infra-Red
Motion note:
These frames approximately follow the global translational motion model.
Related Paper:
First 8 frames of this sequence were used in [1].
Credit:
FLIR research group, Sensors Technology Branch, Wright Laboratory, WPAFB, OH.

Car

Description:
64 uncompressed grayscale images of size 121 x 72
Camera brand:
unknown
Motion note:
These frames approximately follow the global translational motion model + small amount of zoom.
Credit:
Peyman Milanfar

Disk

Description:
26 uncompressed grayscale frames of size 57 x 49
Camera brand:
Olympus C-4000
Motion note:
The first 20 frames approximately follow the global translational motion model. The last 6 have been zoomed in to creat motion outlier.
Credit:
Sina Farsiu

Alpaca (Small)

Description:
55 compressed grayscale frames of size 32 x 70 (These frames were acquired by cropping Alpaca sequence).
Camera brand:
3COM, Model no. 3718
Motion note:
The first 45 frames approximately follow the global translational motion model. The last 10 frames follow a more complicated motion model.
Related Paper:
This sequence was used in [1].
Credit:
Sina Farsiu and Dirk Robinson

Alpaca

Description:
55 compressed grayscale frames of size 96 x 128
Camera brand:
3COM, Model no. 3718
Motion note:
The first 45 frames approximately follow the global translational motion model. The last 10 frames follow a more complicated motion model.
Credit:
Sina Farsiu and Dirk Robinson

Emily (Small)

Description:
A 54 x 35 section of the 53 compressed grayscale frames in the Emily sequence
Camera brand:
3COM, Model no. 3718
Motion note:
These frames approximately follow the translational motion model.
Credit:
Sina Farsiu and Dirk Robinson

Emily

Description:
82 compressed grayscale frames of size 96 x 128
Camera brand:
3COM, Model no. 3718
Motion note:
The first 53 frames approximately follow the translational model. The last 29 frames follow a more complicated motion model.
Credit:
Sina Farsiu and Dirk Robinson

Face 1

Description:
60 compressed grayscale frames of size 38 x 34
Camera brand:
unknown
Motion note:
These frames approximately follow the global translational motion model.
Credit:
Adyoron Intelligent Systems Ltd., Tel Aviv, Israel. Post-processing by Sina Farsiu

Face 2

Description:
40 compressed grayscale frames of size 31 x 32
Camera brand:
unknown
Motion note:
These frames approximately follow the global translational motion model.
Credit:
Adyoron Intelligent Systems Ltd., Tel Aviv, Israel. Post-processing by Sina Farsiu

Surveillance (Rotated)

Description:
20 compressed images of size 76 x 65
Camera brand:
unknown
Motion note:
The first 15 frames approximately follow the Translational model. The last 5 frames follow affine motion model.
Related Paper:
This sequence was used in [1].
Credit:
Adyoron Intelligent Systems Ltd., Tel Aviv, Israel. Post-processing by Sina Farsiu

EIA

Description:
16 low-resolution images in the EIA seqeuence of size 90 x 90. This is a synthesized sequence. These images are acquired by blurring, downsampling, and adding Gaussian noise to the EIA image. Detailed description is available in [1]. Here are the corresponding motion vectors.
Camera brand:
This is a synthesized sequence.
Motion note:
These frames approximately follow the global translational model. Here are the corresponding motion vectors.
Related Paper:
This sequence was used in [1].
Credit:
Sina Farsiu

Full Color Datasets

  • Color Face 1
  • Color Face 2
  • Surveillance (Small)
  • Surveillance
  • Bookcase 1 (Small)
  • Bookcase 1

Color Face 1

Description:
60 compressed color frames of size 38 x 34
Camera brand:
unknown
Motion note:
These frames approximately follow the global translational motion model.
Credit:
Adyoron Intelligent Systems Ltd., Tel Aviv, Israel. Post-processing by Sina Farsiu

Color Face 2

Description:
40 compressed color frames of size 31 x 32
Camera brand:
unknown
Motion note:
These frames approximately follow the global translational motion model.
Credit:
Adyoron Intelligent Systems Ltd., Tel Aviv, Israel. Post-processing by Sina Farsiu

Surveillance (Small)

Description:
40 compressed color images of size 115 x 138
Camera brand:
unknown
Motion note:
These frames approximately follow the global translational motion model.
Related Paper:
This sequence was used in [3].
Credit:
Adyoron Intelligent Systems Ltd., Tel Aviv, Israel. Post-processing by Sina Farsiu

Surveillance

Description:
61 compressed color images of size 288 x 352
Camera brand:
unknown
Motion note:
These frames approximately follow the global translational motion model.
Credit:
Adyoron Intelligent Systems Ltd., Tel Aviv, Israel.

Bookcase 1 (Small)

Description:
30 compressed color frames from Book_case1 seqeuence of size 91 x 121 (these frames were acquired by cropping Book_case1 sequence)
Camera brand:
Pyro 1394
Motion note:
These frames approximately follow the global translational motion model.
Related Paper:
This sequence was used in [3].
Credit:
Sina Farsiu

Bookcase 1

Description:
100 compressed color frames of size 240 x 320
Camera brand:
Pyro 1394
Motion note:
These frames approximately follow the global translational motion model.
Credit:
Sina Farsiu

Raw CFA Demosaicing Datasets

  • Lighthouse
  • Bookcase 2
  • Toy

Bookcase 2

Description:
31 raw CFA frames of size 141 x 181.
Camera brand:
Prototype 2MP CMOS sensor provided by Lior Zimet and Erez Galil from Zoran Corp.
Motion note:
These frames approximately follow the global translational motion model.
CFA Pattern:
The CFA Pattern File
Result:

Kimmel Single Frame Method

Multi-Frame Demosaicing Method
Related Paper:
This sequence was used in [3].
Credit:
Sina Farsiu

Lighthouse

Description:
10 raw CFA frames of size 96 x 64. This is a synthesized sequence. These images are acquired by blurring, downsampling, and adding Gaussian noise to the lighthouse image. Detailed description is available in [3]. Here is the corresponding motion vector.
Camera brand:
This is a synthesized sequence.
Motion note:
These frames approximately follow the global translational motion model. Here is the corresponding motion vector.
CFA Pattern:
The CFA Pattern File
Result:

Kimmel Single Frame Method

Multi-Frame Demosaicing Method
Related Paper:
This sequence was used in [3].
Credit:
Sina Farsiu

Toy

Description:
26 raw CFA frames of size 161 x 261.
Camera brand:
Dragonfly
Motion note:
These frames approximately follow the global translational motion model.
CFA Pattern:
The CFA Pattern File
Result:

Kimmel Single Frame Method

Multi-Frame Demosaicing Method
Credit:
Michael Elad and Eyal Gordon. Post-processing by Sina Farsiu.