FishermanAI/RCap-2025-Fisherman

RCAP 2025 Fisherman

RCAP 2025 Fisherman is a lightweight object detection model optimized for the OpenMV H7 Plus, designed for real-time embedded vision tasks such as robotics, surveillance, and IoT applications.

---

Features

• Ultra-lightweight architecture for OpenMV H7 Plus
• Fast inference and low memory usage
• Multi-class object detection
• Ready-to-run on OpenMV IDE and MicroPython
• Hosted on Hugging Face for easy download

---

Requirements

• Hardware: OpenMV H7 Plus

• Software:

  • OpenMV IDE v3.0+
  • MicroPython for H7 Plus
  • Python 3.10+ (optional, for model conversion/preprocessing)

---

Hugging Face Model

The model is available on Hugging Face: RCAP 2025 Fisherman on HF

You can download it directly in Python:

from huggingface_hub import hf_hub_download

model_path = hf_hub_download(
    repo_id="your-username/rcap-2025-fisherman",
    filename="model.tflite"
)
print("Model downloaded to:", model_path)

Copy the downloaded model.tflite to your OpenMV H7 Plus.

---

Usage on OpenMV H7 Plus

Example MicroPython script:

import sensor, image, time
from rcap_fisherman import RCAPFisherman

sensor.reset()
sensor.set_pixformat(sensor.RGB565)
sensor.set_framesize(sensor.QVGA)
sensor.skip_frames(time=2000)

model = RCAPFisherman("model.tflite")
clock = time.clock()

while True:
    clock.tick()
    img = sensor.snapshot()
    detections = model.detect(img)
    
    for d in detections:
        img.draw_rectangle(d['bbox'])
        img.draw_string(d['bbox'][0], d['bbox'][1]-10, d['label'])
    
    print("FPS:", clock.fps())

---

Training / Fine-Tuning

1. Prepare your dataset in COCO or Pascal VOC format.
2. Train using PyTorch or TensorFlow scripts.
3. Convert your model to TFLite for deployment:

python convert_to_tflite.py --model_path trained_model.h5 --output_path model.tflite

4. Upload the .tflite model to your OpenMV H7 Plus.

---

License

This project is licensed under the Apache License 2.0. See LICENSE for details.