Seabream Freshness Dataset
About:
The
Seabream freshness dataset documents the decaying process of 40 different
seabream (sparus aurata) fish, documented for a time span of up to 14 days.
Images
were acquired on two different locations: (i) at the Universidade de Aveiro, by
Renato Pinto, in 2020; and (ii) in a cooperation between Instituto de
Telecomunicacoes – Instituto Superior Tecnico, Universidade de Lisboa and
Universidade de Aveiro, by João Rodrigues, in 2023, capturing images of fishes
25 to 40.
The
first imaging campaign, in 2020, consisted of 10 photograph sessions, capturing
images of fishes 1 to 24. A time span of 7 days (168 hours) was considered,
with the first 4 days including 9 sessions, at 0, 17, 24, 41, 48, 65, 72, 89
and 96 hours after the fish capture, plus a final session after 168 hours (7
days). Each session includes 6 pictures of each fish (with eventual
duplicates). Images were captured using a smartphone (Samsung with SM-N975F
camera) and a digital camera (Canon EOS 800D). Images were captured from
various angles, with and without special focus of the eye region, and
considering two different types of illumination, notably neutral and yellow
saturated.
The
second imaging campaign, in 2023, consisted of 22 photograph sessions,
capturing images of fishes 25 to 40. A time span of 14 days (326 hours) was
considered. With sessions approximately spaced by 12h in the first 9 days, and
a spacing of 24h afterwards. Images were captured using a smartphone (Samsung
with a SM-A226B camera). Each session includes 8 images of each fish, 4 from
each side of the fish, from various angles, focusing both in
the eye-region as well as the whole fish. Each session also contained group
pictures of 4 different fishes, which were rotated in each session.
The
annotation process was carried out using the Segments AI online platform
(segments.ai), labelling the eye-region, as well as the whole fish region.
Details:
The
database is organized in the following hierarchical folder structure:
·
The
main directory, “seabream_freshness”, contains 2 folders: (i)
”original imgs”, corresponding to the original images taken in the
sessions; and (ii) ”eye segmented imgs”, corresponding to the eye region, which
was segmented from the original images. Each of these 2 folders contains 28
subfolders corresponding to time elapsed since the fish capture;
·
Each
of the 28 folders contains folders corresponding to the images of each
individual fish (1, ..., 40), and group photos (G_X1_X2_X3_X4, where X1, ...,
X4 are the numbers of the fishes in the image);
·
The
following and final layer of folders corresponds to each fish image folder and
has as its name the corresponding fish number plus the acquisition date,
whenever images were acquired using a smartphone. For images acquired using the
digital camera the file name is the fish number followed by IMG_XXXX, where
XXXX is the sequential number of the photos taken by the digital camera;
·
For
every fish image folder there are 2 types of files: (i) the .JPEG photo of the
fish; and (ii) the corresponding ground truth segmentation image, stored as a
color image in PNG color format (fish eye – orange;
fish body – yellow; background – black).
License agreement:
· To use the database please fill in
the license agreement and send a
scanned copy of the signed form by e-mail.
Download:
· The database can be downloaded from
the links below (the size of each part is between 1 and 2 GB).
· A password for decrypting the compressed
Zip files will be provided after receiving the duly signed license agreement
(see above).
How to reference the database:
· J. Rodrigues, O. Pacheco, and P.L.
Correia, “Seabream Freshness Classification using Vision Transformers”, Proceedings
of the Iberoamerican Congress on Pattern Recognition - CIARP, Coimbra, Portugal,
Nov. 2023
Contacts:
Feedback is
welcome. Please provide your comments and suggestions by e-mail.