sandbox338/inception-wildlife-detector-detectron2

Wildlife Detector - Detectron2

A Faster R-CNN model with optimized Inception v1 backbone for detecting 10 wildlife species.

Classes

• Antelope
• Buffalo
• Elephant
• Giraffe
• Gorilla
• Leopard
• Lion
• Rhino
• Wolf
• Zebra

Performance

Metric	Value
AP	45.7%
AP50	81.8%
AP75	47.7%

Per-Class Performance (AP)

Animal	AP	Animal	AP
Buffalo	58.9%	Elephant	58.5%
Gorilla	51.2%	Leopard	49.4%
Wolf	48.0%	Antelope	46.4%
Rhino	44.1%	Zebra	43.7%
Lion	31.9%	Giraffe	24.5%

Usage

import torch
from detectron2.config import get_cfg
from detectron2.engine import DefaultPredictor
from detectron2 import model_zoo
from huggingface_hub import hf_hub_download

# Download model
model_path = hf_hub_download(
    repo_id="mynane/inception-wildlife-detector-detectron2", 
    filename="pytorch_model.bin"
)

# Setup config
cfg = get_cfg()
cfg.merge_from_file(model_zoo.get_config_file("COCO-Detection/faster_rcnn_R_50_C4_3x.yaml"))
cfg.MODEL.WEIGHTS = model_path
cfg.MODEL.ROI_HEADS.NUM_CLASSES = 10
cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.5

# Create predictor
predictor = DefaultPredictor(cfg)

# Inference
import cv2
image = cv2.imread("wildlife_image.jpg")
outputs = predictor(image)

Model Details

• Architecture: Faster R-CNN with Optimized Inception v1 backbone
• Framework: Detectron2
• Input Size: 800x1333 (min x max)
• Confidence Threshold: 0.5