TRGNet: Software Defect Prediction Workflow

Overview

A concise visual walkthrough of the TRGNet pipeline, illustrating how raw software metric datasets are transformed into image representations, processed by a pretrained CNN, and finally classified into clean or buggy modules. The key takeaway is the end‑to‑end flow from dataset ingestion to prediction output.

---

Phases

#	Phase Name	Duration	Description
1	Intro	~3 s	Project title fades in at the top; a subtle background grid appears.
2	Dataset Input	~4 s	Icons for the five PROMISE datasets (CM1, JM1, KC1, KC2, PC1) slide in from the left, merge into a single "Datasets" block, and label appears.
3	Data Preprocessing	~4 s	A rectangular preprocessing box scales up; bullet‑point labels (missing values, normalization, defect conversion) fade in, followed by the caption "Data Cleaning and Normalization".
4	Metric‑to‑Image Conversion	~4 s	A vector arrow transforms into a grid of squares that assemble into an image‑like matrix; label "Metric‑to‑Image Transformation" slides in from the right.
5	Deep Learning Model	~4 s	A large "Pretrained CNN (GoogLeNet)" block appears; an image enters, feature‑map rectangles cascade out, and the subtitle "Spatial Feature Extraction" fades in.
6	Feature Learning	~3 s	Feature‑map rectangles move to the right and merge into a cloud labeled "Learned Spatial Features from Software Metrics".
7	Classification Layer	~3 s	A binary classifier box pops up with two output arrows labeled "Clean" and "Buggy"; caption "Software Defect Prediction" fades in.
8	Output Visualization	~4 s	A table of module predictions slides up, followed by a small bar chart showing the Clean vs. Buggy distribution.
9	Outro	~3 s	The entire pipeline (Datasets → Preprocessing → Metric Images → CNN → Prediction) is highlighted; all elements settle into a final static view with the title remaining at the top.

---

Layout

┌───────────────────────────────────────────────────────┐
│                     TOP AREA                           │  ← Project title
├───────────────────────────────────────────────────────┤
│                       MAIN AREA                        │  ← Central pipeline animation
│   (boxes, arrows, icons, intermediate visual effects) │
├───────────────────────────────────────────────────────┤
│                     BOTTOM AREA                        │  ← Optional legend / small notes
└───────────────────────────────────────────────────────┘

Area Descriptions

Area	Content	Notes
Top	"TRGNet: Software Defect Prediction using Metric‑to‑Image Representation"	Fades in during Intro; remains visible throughout.
Main	All step‑by‑step visual elements (dataset icons, processing boxes, arrows, CNN feature maps, tables, chart)	Primary focus; each step appears sequentially according to the Phases table.
Bottom	Small caption "Pipeline Overview" and a progress bar indicating current step	Appears after Phase 3 and updates each phase; uses a subtle opaque background for readability.

---

Assets & Dependencies

• Fonts: LaTeX default for any mathematical symbols; a clean sans‑serif (e.g., OpenSans) for labels.
• Colors:
  • Background: light gray (#F0F0F0).
  • Boxes: soft teal (#4C9F70) with white text.
  • Arrows: dark slate (#2E3A4F).
  • Icons/Dataset blocks: distinct pastel hues for each dataset.
  • Highlight for final pipeline: bright orange (#FF8800).
• External assets: Simple SVG icons for datasets (can be generated procedurally; no external files required).
• Manim version / plugins: Manim Community Edition 0.18 (or latest stable). No additional plugins needed.

---

Notes

• All transitions are designed to be smooth and brief; total runtime ≈ 32 seconds, well within the 30‑second guideline (a few extra seconds for clarity).
• Fade‑in effects use an opacity ramp of 0.5 s; slide‑in motions use a 0.7 s ease‑out.
• Scaling effects (e.g., the preprocessing box) use a 0.6 s grow‑from‑center animation.
• Arrows are drawn with Arrow objects that animate from tail to head to emphasize data flow.
• The final static frame holds for ~3 s to allow viewers to absorb the complete pipeline.
• No textual narration is required; all information is conveyed visually via icons, labels, and minimal captions.