MIRA

Open computer vision research

Computer vision for recycling.

MIRA detects five material classes, measures uncertainty, and prepares every result for edge deployment.

View the research Run MIRA locally
MIRA dashboard Local system
MIRA local dashboard with camera area, detection statistics, and class distribution panels
Real dashboard interface included in the repository.
90.58%EXP-019 mAP50
5material classes
19recorded experiments

What the current system can do.

I can capture a frame, detect the materials, inspect the confidence, and review the result in one local workflow.

  1. Capture

    USB and IP cameras feed a thread-safe frame buffer with reconnect and freeze detection.

  2. Detect

    YOLO or TFLite models locate glass, metal, paper, plastic, and trash.

  3. Decide

    Confidence and reject thresholds separate confident detections from uncertain ones.

  4. Inspect

    The dashboard reports detections, class counts, latency, FPS, CPU, and memory.

I built the parts around the model too.

MIRA includes the tools I needed to prepare data, compare runs, test a camera stream, and inspect the result.

Live vision

Local inference with a real operating view.

Model selection, camera controls, bounding boxes, class distribution, recent detections, and system metrics are joined in one FastAPI dashboard.

Dashboard reference
Research 19

documented experiments

Classification, detection, dataset composition, quantization, and repeatability.

Data

Deterministic dataset tools.

Merge registered sources, remap labels, validate YOLO annotations, generate manifests, and reproduce splits.

The data changed the result more than the model did.

EXP-018 and EXP-019 reached about 90.6% mAP50 after I removed inconsistent examples and rebuilt the detector around a cleaner dataset.

mAP50-95
82.15%
Precision
87.2%
Recall
84.6%
Training time
2.67 h
See the experiment record
Stage B · mAP50

Detector progression

90.58%
Current through EXP-019. Values are generated from the repository experiment log.

Here is what works, and what still needs testing.

Working now

  • Five-class object detection
  • Live USB and IP camera inference
  • Confidence reject system
  • Training and dataset pipeline
  • Dashboard and model exports

Still to validate

  • Raspberry Pi latency and memory
  • End-to-end physical sorting
  • Robustness under heavy occlusion
  • More examples of crumpled paper and end-on cans

Start with the results or the code.

Research record19 experiments DocumentationSetup to API Source codeGitHub repository