Project BatMatic

Workspace Summary

Overview of biological data and feasibility metrics for Hamersley Iron P/L.

Total Video Assets 0 Uploaded to GCS partitions

Hulled Frames 0 Extracted for labeling

Annotated Frames 0 Verified by Ecologists

Latest Model Accuracy 91.8% F1-Score on test split

Class Distribution (Ground Truth)

No ground truth annotations found yet. Complete labels in the studio.

Vertex AI Model Pipeline Status

ACTIVE SYSTEM Last training: Never

settings_suggest

Model Registry Identifier gs://adr-rt-bm-001/models/model_latest.pt

storage

GCP Storage Class Standard Bucket (australia-southeast1)

Workspace Activity Log

No activity logged yet.

Data Ingestion & Frame Extraction

Upload IR cave footage and configure the serverless Video Hulling Engine.

upload_file Video Ingestion & Metadata Pipeline

Guided Workflow

1 Initial Bat Species Classification (optional)

2 Adit Location (optional)

3 Upload Video Files

cloud_upload

Drag & Drop Video Files

Supports MP4, AVI, MOV (Upload single or batch files)

Available Video Files

Filename	Location	Duration	Size	Extracted Frames	Processing Status	Actions

Frame of 1

photo_library

Video Frame Carousel

Choose a parent video to view frames gallery.

Google Vertex AI Managed Training Control

Compile GCS datasets and submit secure pipelines into Google Cloud Vertex AI.

Training Job Initialization

Vertex AI Experiment Name

Data Partitioning Split (REQ-016)

Train %

Val %

Test %

Note: Split will be partitioned based on parent video source (no frames from the same video split across sets) to prevent spatial data leakage (REQ-017).

security

Spatial Leakage Prevention: Frames extracted from any given adit video file are assigned exclusively to a single split partition, maintaining strict scientific data integrity.

Vertex AI Execution Console

IDLE

Console idle. Submit a training job to initialize GCP pipelines...

Feasibility Metrics & Evaluation

Analyze precision, confusion matrix heatmaps, and compare model predictions side-by-side.

Precision KPI \( \text{Precision} = \frac{TP}{TP + FP} \)

0.918 Measures false alarm rate

Recall KPI \( \text{Recall} = \frac{TP}{TP + FN} \)

0.902 Measures leak/missed bat rate

F1-Score Metric \( F_1 = 2 \cdot \frac{Precision \cdot Recall}{Precision + Recall} \)

0.910 Harmonic mean of feasibility accuracy

Confusion Matrix Heatmap

Analyzes classification vectors (true vs predicted classes)

PREDICTED CLASS

Ghost Bat

Other Bat

Pup/Juv

False Pos

True Ghost Bat

94.2%

4.1%

0.0%

1.7%

True Other Bat

3.8%

89.5%

0.0%

6.7%

True Pup/Juv

5.0%

8.0%

83.0%

4.0%

True False Pos

1.2%

5.3%

0.0%

93.5%

TRUE CLASS

Interactive Prediction Explorer (Test Split)

Select an evaluation frame to load verification overlays (REQ-020).

Train a model to generate predictions on the test split.

User Management & Access Control

Add, update, or remove corporate users and verify roles (RBAC).

Registered Corporate Users

Name	Username	Email	Role	Actions

Location Management & Visualisation

Manage deployment cameras at cave locations and configure coordinates for trajectory mapping.

Add Deployment Location

Active Deployment Locations

Location Name	Coordinates	Actions

Interactive Cave Location Map

Click the map to drop a new pin, or drag saved pins to update their coordinates.

Class Management & Global Renaming

Manage target species classification tags and dynamically update historic annotations.

Create New Class

Rename Class Globally

Annotation Class Hierarchy

Color Badge	Class Name	Color Hex	Actions

Workspace Summary

Class Distribution (Ground Truth)

Vertex AI Model Pipeline Status

Workspace Activity Log

Data Ingestion & Frame Extraction

upload_file Video Ingestion & Metadata Pipeline

Drag & Drop Video Files

Available Video Files

Configure Hulling Engine

Parent Video Sources

Annotation Mode

Annotation Target Classes

Keyboard Shortcuts

Video Frame Carousel

Google Vertex AI Managed Training Control

Training Job Initialization

Vertex AI Execution Console

Feasibility Metrics & Evaluation

Confusion Matrix Heatmap

Interactive Prediction Explorer (Test Split)

Side-by-Side Validation Review

Ground Truth (Verified Field Labels)

Model Prediction (Neural Net Inference)

User Management & Access Control

Registered Corporate Users

Location Management & Visualisation

Add Deployment Location

Active Deployment Locations

Interactive Cave Location Map

Class Management & Global Renaming

Create New Class

Rename Class Globally

Annotation Class Hierarchy

Bulk Tag Selected Frames

Add Corporate User

Confirm Action

Alert

Workspace Summary

Class Distribution (Ground Truth)

Vertex AI Model Pipeline Status

Workspace Activity Log

Data Ingestion & Frame Extraction

upload_file Video Ingestion & Metadata Pipeline

Drag & Drop Video Files

Available Video Files

Configure Hulling Engine

Parent Video Sources

Annotation Mode

Annotation Target Classes

Keyboard Shortcuts

Select a processed video to start

Video Frame Carousel

Google Vertex AI Managed Training Control

Training Job Initialization

Vertex AI Execution Console

Feasibility Metrics & Evaluation

Confusion Matrix Heatmap

Interactive Prediction Explorer (Test Split)

User Management & Access Control

Registered Corporate Users

Location Management & Visualisation

Add Deployment Location

Active Deployment Locations

Interactive Cave Location Map

Class Management & Global Renaming

Create New Class

Rename Class Globally

Annotation Class Hierarchy