How muhammad-zainal-muttaqin/research-pipeline works
Research-Pipeline combines a Scopus literature-review workflow, computer-vision experiments, and a documented four-channel RGB-depth deployment pipeline. Researchers use the literature tools to build and audit review evidence, and use experiment training and evaluation scripts to test detection approaches; the README identifies RF-DETR-L on E-021 as the citeable four-class result. The production pipeline prepares depth, trains a shared RGB/RGB-depth model, and runs inference. Evidence is limited to the README, tree, and sampled excerpts; unsampled subsystems are represented only at the documented directory level.
Main entry points: tools Review tooling · train Model training · fourch.py Four-channel core · train_fusion_2branch.py Two-branch fusion
Components 17
Literature review
- Review tooling
tools - Scopus workflow
tools/scopus - Evidence synthesis
tools/build_evidence_matrix.py - Screening tools
tools/build_literature_screening_master.py
Detection experiments
- Model training
experiments/code/train - RF-DETR baseline
experiments/code/train/train_rfdetr.py - RF-DETR variants
experiments/code/train/train_rfdetr_freq.py - Two-branch fusion
experiments/code/train/train_fusion_2branch.py - COCO evaluation
experiments/code/eval/eval_all_pycoco.py - Paired E-022 evaluation
experiments/code/eval/eval_e022_paired.py - Metric diagnostics
experiments/code/eval/eval_extras.py
Research analysis
- Experiment analysis
experiments/code/analysis
Depth deployment
- Depth preparation
pipeline/prepare_depth.py - Four-channel core
pipeline/fourch.py - RGB-depth training
pipeline/train_4ch.py - Field inference
pipeline/infer_4ch.py
Other components
- Researcher
Connections 18
- Researcher → Review tooling: builds review
README.md - Review tooling → Scopus workflow: uses corpus tools
README.md - Review tooling → Screening tools: screens literature
README.md - Review tooling → Evidence synthesis: synthesizes evidence
README.md - Researcher → Model training: runs experiments
README.md - Model training → RF-DETR baseline: includes baseline
experiments/code/train/train_rfdetr.py - Model training → RF-DETR variants: includes variants
experiments/code/train/train_rfdetr_freq.py - Model training → Two-branch fusion: includes fusion
experiments/code/train/train_fusion_2branch.py - COCO evaluation → Metric diagnostics: provides metrics
experiments/code/eval/eval_extras.py - Paired E-022 evaluation → COCO evaluation: imports evaluator
experiments/code/eval/eval_e022_paired.py - Paired E-022 evaluation → Two-branch fusion: loads fusion classes
experiments/code/eval/eval_e022_paired.py - Two-branch fusion → Four-channel core: imports core
experiments/code/train/train_fusion_2branch.py - Researcher → Depth preparation: prepares sensor data
README.md - Depth preparation → Four-channel core: encodes depth
pipeline/prepare_depth.py - Researcher → RGB-depth training: trains deployment model
README.md - RGB-depth training → Four-channel core: configures training
pipeline/train_4ch.py - Researcher → Field inference: runs field inference
README.md - Field inference → Four-channel core: predicts detections
pipeline/infer_4ch.py