Table 1 Brief comparison of SCSFish2025 with other relevant datasets mentioned above.
From: SCSFish2025: a large dataset from South China sea for coral reef fish identification
Dataset name | Resolution | Number of species | Number of annotations | Data source | Annotation type | Environment type | Key features and limitations |
|---|---|---|---|---|---|---|---|
FishCLEF 2015 | 320*240 / 640*480 | 15 | 9162 | Fish4Knowledge video resource library (coral reefs in Taiwan, China) | Bounding box | Natural marine environment | Low resolution; corruption; label duplication |
F4K with Complex Scenes | 320*240 / 640*480 | Not specified | 3500 | Fish4Knowledge video resource library (coral reefs in Taiwan, China) | Segmentation Mask | Natural marine environment | For background modeling and foreground detection |
A Large Scale Fish Dataset | 590*445 | 9 | 9000 | Seafood from supermarket | Category | Controlled environment | Lighting conditions are carefully adjusted; background remains monotonous |
QUTFish | 480*360 | 468 | 3960 | Not specified | Bounding box | Control; off water; in place | Only one bounding box per image; most images taken in a controlled environment |
F4K with recognition | 352*240 | 23 | 27,370 | Fish4Knowledge video resource library (coral reefs in Taiwan, China) | Category | Natural marine environment | Just for classification; low resolution |
WildFish | various | 1000 | 54,459 | Web Scraping | Category | Natural marine environment | Largest image dataset for wild fish identification; for classification only |
Croation Fish | various | 12 | 794 | Adriatic Sea in Croatia | Category | Natural marine environment | Each image block described by bounding boxes is extracted and saved as individual dataset images |
OzFish | 1920*1080 | 507 | 45,000 | Baited Remote Underwater Video Stations | Bounding box | Natural marine environment | Bounding Box Species Annotations; Bounding Box Tail Annotations; Labels are for fish/non-fish only |
DeepFish | 1920*1080 | 2 | 19,766 for classification, 3200 for detection, 620 for segmentation | Mostly in North Eastern Australia | Category; Bounding box; Segmentation mask | Natural marine environment | Addresses 4 tasks: lassification, counting, detection, and segmentation; for detection tasks, the annotation is fish/non-fish only |
SEAMAPD21 | Not specified | 130 | 90,000 | Gulf of Mexico | Bounding box | Natural marine environment | Includes stereo vision (can be used for 3D object detection); categories are unbalanced, with occlusion in front of the shooting equipment |
SCSFish2025 | 1920*1080 | 30 | 120,084 | South China Sea | Bounding box | Natural marine environment | High-resolution; fish species-rich; well-labeled |
