Addax-Data-Science/Japan_Gifu_v0.2

Original repo: https://github.com/gifu-wildlife/MDetToolsForJCameraTraps

The model weights have been soptred here for easy integration with AddaxAI: https://addaxdatascience.com/addaxai/

Gifu-wildlife cls model v0.2 (resnet50) for AddaxAI v6.14

Filename (weight): gifu-wildlife_cls_resnet50_v0.2.1.pth

Changes from v0.1:

• Trained using data excluding “human” and “dog” due to insufficient data. As a result, the number of categories decreased from 15 to 13.
• Modified classify_datections.py to match changes in AddaxAI v6.14.
• Changed “rabbit” to “hare.”
• Carefully rewrote classes.csv.

Test results[2025-05-20]:

• Top1 Accuracy : 0.926
• Top3 Accuracy : 0.987
• class wise acc | bear:0.816 | bird:0.571 | boar:0.941 | cat:0.200 | deer:0.971 | fox:0.607 | masked:0.519 | monkey:0.000 | mustelidae:0.600 | rabbit:0.235 | raccoondog:0.871 | serow:0.884 | squirrel:0.649

Note: It should be noted that the dataset is biased toward deer, which may lead to an overestimation of both Top1 and Top3 accuracy.