Smart Target Analysis & Feedback System

STAFS
PRECISION
VISION
PLATFORM

An end-to-end shooting feedback platform built for real-time tracking, MPI zeroing analysis, collaborative annotation, and a continuously improving data pipeline for performance optimization.

3
Detection Classes
2
Model Variants
AR
Nano Ready
DETECTING • BULLS_EYE
CONF: 0.94 • MPI LOCKED
📷 IMAGE CAPTURED
⚙️ BACKEND INFERENCE
✓ FEEDBACK READY
01 — System Architecture

The Complete Closed Loop

Every interaction feeds the system — from the first photo taken to the trained model update.

📷
Photo / Upload
☁️
Backend API
🧠
AI Inference
🎯
Detection Result
📊
User Feedback
✏️
Notation & Save
🔄
Training Queue
Continuous Training
🚀
Model Upgrade
02 — User Journey

From Shot to Insight

A seamless interaction loop that captures data, delivers real-time AI feedback, and routes everything back into model improvement.

📷

Capture — Photo or Upload

The user photographs a physical target directly through the app camera or uploads an existing image from their gallery. High-resolution inputs are accepted and pre-processed automatically for model inference.

Camera Capture Gallery Upload Auto Pre-process
☁️

Transmission to Backend

The image is securely transmitted to the backend server. A RESTful API layer manages request queuing, authentication, and efficient forwarding to the inference engine — ensuring fast turnaround even under high load.

REST API Secure Transfer Request Queue
🧠

AI Model Inference

The STAFS detection model processes the image, identifying and localizing all relevant objects. The model returns bounding boxes, class labels, and confidence scores for each detected element: bullet holes, aim assist zones, and bulls-eye positions.

Object Detection Bounding Boxes Confidence Scoring 3 Class Detection
🎯

MPI Zeroing Analysis

Using detected bullet hole positions, the system computes the Mean Point of Impact (MPI). The MPI is compared against the bulls-eye center to derive zeroing offset — a precise vector indicating how much to adjust the sight in any direction to achieve zero at that distance.

MPI Computation Offset Vector Zero Alignment
📊

User Feedback Delivery

Results are displayed immediately in the app: an annotated target overlay with detected classes highlighted, grouped scatter analysis, MPI marker, and actionable zeroing guidance. Feedback is visual, intuitive, and instantly interpretable.

Visual Overlay Scatter Analysis Zeroing Guidance
✏️

Notation & Session Save

Users can annotate results with notes — distance, weather, load type, stance — before saving the session. All raw images, detections, and metadata are stored. Annotated corrections made by users are flagged as verified ground truth.

Session Notes Metadata Store Ground Truth Flagging
🔄

Training Queue Submission

Saved sessions — particularly those with user-verified annotations — are automatically pushed into a training data queue. A background pipeline processes and formats entries, ensuring clean, balanced datasets for the next training cycle.

Auto Queuing Dataset Formatting Verified Ground Truth
🚀

Continuous Training & Model Upgrade

The model is continuously fine-tuned on accumulated real-world data. When performance metrics cross defined thresholds, the new model version is promoted to production — replacing the previous deployment. The system self-improves with every user interaction.

Continuous Learning Version Promotion Auto Deployment Performance Gating
03 — Detection Models

Dual Model Architecture

Two specialized model variants — one optimized for accuracy in full analysis sessions, one ultralight for real-time AR use.

Production
V4 — Nano
Ultralight model optimized for real-time AR use with minimal latency and resource requirements.
Precision
97.7%
Recall
94.8%
mAP50
96.2%
mAP50-95
76.7%
Baseline
V3 — Fast
High accuracy model for full analysis sessions with comprehensive detection capabilities.
Precision
97.7%
Recall
95.2%
mAP50
95.8%
mAP50-95
78.9%
Detection Classes — Dataset Config
path: ./
train: images/train
val: images/val
nc: 3
names: ['bullet', 'aim_assist', 'bulls_eye']
CLASS 0
Bullet
Individual bullet hole impact points on the target surface. Each detected hole contributes to the MPI calculation and scatter group analysis.
Impact Point MPI Input
CLASS 1
Aim Assist
Reference bounding region defining the shooter's intended aim zone. Provides spatial context for grouping and relative accuracy calculation.
Aim Zone Spatial Ref
CLASS 2
Bulls Eye
Central scoring zone — the primary reference for zeroing offset computation. The geometric center of this detection is used as the ideal impact point.
Zero Reference Score Center
04 — Ballistic Intelligence

MPI Zeroing Engine

Automatically computes the Mean Point of Impact from detected bullet hits and derives the precise zeroing correction vector.

MPI — COMPUTED
ZEROED GROUP

Automated Sight Correction Intelligence

After a session, the system aggregates all detected bullet impacts. The centroid of these points forms the Mean Point of Impact — which is then compared to the detected bulls-eye center to compute the required sight adjustment.

The output is a concrete, actionable vector: the exact number of clicks up/down and left/right to achieve perfect zero at the given distance.

01

Bullet Detection

All class-0 (bullet) detections are extracted from the inference result. Each centroid coordinate is recorded.

02

MPI Calculation

The arithmetic mean of all bullet centroids gives the Mean Point of Impact — the geometric center of the group.

03

Bulls-Eye Reference

The centroid of the class-2 (bulls_eye) detection provides the ideal zero reference point on-target.

04

Offset Vector & Correction Output

The delta between MPI and bulls-eye is converted to sight adjustment units and displayed as clear directional guidance to the user.

05 — Data Quality

Notation & Ground Truth

Users don't just receive feedback — they actively contribute to model quality through built-in annotation and session notation tools.

aim_assist
bullet
bulls_eye
bullet
bullet
DETECTIONS: 5 objects found
STATUS: ANNOTATED ✓

Session Notation System

After reviewing AI detections, users can add session context — distance, conditions, corrections — and flag any misdetections for manual review. This user-verified data becomes priority training material.

📝

Session Notes

Attach distance, weather, ammunition type, and stance data directly to each session record for richer training context.

Detection Verification

Users confirm or correct AI detections. Verified sessions are flagged as high-confidence ground truth in the training queue.

🗂️

Persistent Session History

All sessions are saved with full metadata, detection overlays, and zeroing reports — accessible anytime for longitudinal progress tracking.

🔁

Auto Training Submission

Saved and verified sessions are automatically queued to the training pipeline — no manual data management required.

06 — Self-Improvement Engine

Continuous Training Loop

The model never stops learning. Every session generates data that feeds back into the pipeline, continuously elevating accuracy and generalization.

USER SESSION TRAINING QUEUE FINE TUNE VERSION PROMOTE DEPLOY MODEL AI FEED BACK STAFS LEARNING LOOP

Perpetual Model Evolution

Unlike static deployments, STAFS operates as a living system. Every annotated session becomes training data. Every training run has the opportunity to produce a better model — which is then gated, tested, and automatically promoted if it outperforms the current version.

baseline
V3 — Fast
Initial trained model. Established detection capability for all 3 classes. Served as production baseline.
current production
V4 — Fast
Trained on expanded real-world dataset. Significant precision and recall improvements. Now the primary inference engine.
in training
V5 — In Queue
Actively accumulating annotated session data. Will undergo evaluation before promotion decision.
nano variant
Nano — AR Build
Compact model variant optimized for on-device AR inference. Runs on constrained hardware with real-time throughput.
07 — Extended Reality

Nano Model & AR Integration

A purpose-built ultralight model variant powers real-time augmented reality overlays directly on device — zero cloud latency.

🥽

Augmented Reality Overlay

The Nano model variant is designed to run fully on-device within AR environments. It overlays real-time detection results — bullet impacts, aim zones, bulls-eye markers — directly onto the physical target as seen through the AR camera.

This eliminates the round-trip to the backend for real-time feedback, enabling sub-50ms response at the device level. Perfect for live training scenarios where latency is critical.

On-Device Inference Real-Time Overlay Sub-50ms Latency No Cloud Required
Model Variant Nano (Ultralight)
Deployment Target AR Headset / Mobile AR
Detection Classes 3 (bullet, aim_assist, bulls_eye)
Inference Mode On-Device (no internet)
Overlay Type Real-Time Bounding + MPI
Model Source Distilled from Fast Variant

Edge-Optimized Architecture

The Nano variant undergoes quantization and architectural pruning derived from the full Fast model. It retains detection quality for all three target classes while dramatically reducing parameter count to fit within AR hardware memory budgets.

Quantized Weights Pruned Architecture Low Power Draw
🎯

Live MPI in AR Space

Beyond simple detection, the AR module computes and renders the MPI vector live — projecting the calculated group center and zeroing offset directly onto the physical target plane in real time, visible through the AR lens.

Live MPI Overlay 3D Projection Target Plane Aware
08 — Metrics & Benchmarks

Detection Performance

Quantified model metrics across both deployed versions and per-class breakdown.

Per-Class Performance — V4 Nano
Class Precision Recall mAP50
Bullet
89%
86%
88%
Aim Assist
95%
93%
94%
Bulls Eye
90%
87%
89%
Overall 97.7% 94.8% 96.2%
Version Comparison
Version mAP50 mAP50-95 Status
V4 Nano 96.2% 76.7% Active
V3 Fast 95.8% 78.9% Baseline
Training Progress — mAP50 over Epochs
0 0.5 0.75 1.0 V4 96% V3 96%
Improvement V3 → V4
Precision0%
Recall-0.4%
mAP50+0.4%
mAP50-95-2.2%