How zable-star/imaging works
This is an auditable research prototype for simulated laser range-gated imaging, multi-gate 3D-object and rendering-authenticity classification, and conditional cross-domain transfer. The principal workflow constructs gated-image manifests, trains a ModelNet5 source model, transfers compatible weights to military-object folds, and summarizes out-of-fold results. The README documents Blender rendering and formal evaluation; where their implementation files are not in the supplied tree, the graph uses the existing enclosing scripts/ path. Other capabilities are shown as separate branches; unsampled wiring is omitted.
Main entry points: build_modelnet_multitask_manifest.py Multitask manifests · run_multitask_transfer_cv.py Military transfer folds · multitask_dataset.py Multitask data loader · run_modelnet_scheme3.py ModelNet5 pretraining
Components 23
Data and Simulation
- Pair preparation
scripts/prepare_modelnet_scheme3_pairs.py - Rendering and data build
scripts - Military data curation
dataset_new/generate.py - Manual dataset review
dataset_new/review_dataset.py - Multitask manifests
scripts/build_modelnet_multitask_manifest.py
Models and Training
- Multitask data loader
multitask_dataset.py - ModelNet manifest loader
modelnet_manifest_dataset.py - Dual-task network
multitask_model.py - Source training
scripts/train_multitask_manifest.py - Slice attention classifier
model.py - Baseline training
train.py - Experiment runner
run_experiments.py
Transfer and Evaluation
- ModelNet5 pretraining
scripts/run_modelnet_scheme3.py - Military transfer folds
scripts/run_multitask_transfer_cv.py - Evaluation protocol
evaluation_protocol.py - Formal results summary
scripts/summarize_triangle_transfer_formal.py - OOF results and figures
scripts
Research Analysis
- Detection prototype
detection_model.py - Gate dataset diagnostics
dataset_new/diagnose_gate_stack.py - Attention analysis
analyze_attention_scores.py - Gate sparsity analysis
analyze_gate_sparsity.py
Other components
- Researcher
- Blender rendering
Connections 28
- Researcher → Pair preparation: prepares pairs
- Researcher → Military data curation: curates objects
- Researcher → Rendering and data build: requests rendering
- Pair preparation → Blender rendering: supplies pairs
- Blender rendering → Rendering and data build: renders gates
- Rendering and data build → Multitask manifests: provides image data
- Military data curation → Manual dataset review: reviews candidates
- Manual dataset review → Multitask manifests: informs selection
- Multitask manifests → Multitask data loader: loads samples
- Multitask manifests → ModelNet manifest loader: loads source samples
- Multitask data loader → Source training: batches samples
- Multitask data loader → Military transfer folds: batches target folds
- ModelNet manifest loader → ModelNet5 pretraining: loads source data
- ModelNet5 pretraining → Source training: runs pretraining
- Source training → Dual-task network: optimizes
- ModelNet5 pretraining → Military transfer folds: provides checkpoint
- Military transfer folds → Dual-task network: trains target tasks
- Military transfer folds → Evaluation protocol: evaluates folds
- Evaluation protocol → Formal results summary: supplies metrics
- Formal results summary → OOF results and figures: summarizes OOF
- Researcher → Baseline training: runs baseline
- Baseline training → Slice attention classifier: trains classifier
- Experiment runner → Baseline training: dispatches runs
- Researcher → Experiment runner: configures experiments
- Researcher → Detection prototype: explores detection
- Researcher → Gate dataset diagnostics: audits gate data
- Researcher → Attention analysis: analyzes attention
- Researcher → Gate sparsity analysis: analyzes sparsity