Table 1 Summary of previous datasets found in the literature, and comparison to proposed dataset.

From: The DeepFish computer vision dataset for fish instance segmentation, classification, and size estimation

Dataset

Images (instances)

Categories

Aim of dataset

Spampinato et al.5

360 (360)

10

classification

Ogunlana et al.6

150 (150)

2

classification

Cutter et al.1

929 (1,005)

detection (underwater)

French et al.4

443 (–)

counting (on trawler belts)

Sung et al.2

video data

detection (underwater)

fish4Knowledge11

2.5·106 to 16·106 fish

24

detection, classification (underwater)

NCFM10

8

classification

Chhabra et al.8

435 (435)

8

classification (underwater)

Alsmadi et al.9

400 (400)

24

classification

Fish-Pak7

915 (915)

6

classification

Zhang et al.3

1,501 (150k)

detection, counting

Brackish dataset12

89 videos

6

detection, classification

URPC18

5,543 (41,441)

5

classification (underwater)

DeepFish (Proposed)

1,291 (7,339)

59

instance segmentation, classification, size estimation (at fish market)

  1. The ground truth available for each is limited by the ‘aim of the dataset’ (i.e. classification datasets do not provide pixel-wise masks).