Simulation Dataset
4th Koh Young AI Contest — ICCAS 2026
📘 This is a supplementary document. For official contest rules, see Contest Rules ★
1 Dataset Overview
Each scenario folder contains the PCB component layout, equipment parameters, and step region information. The solver reads these files to optimize FOV placement.
📁 Folder Structure
simulation_data/
S00_sparse_fast_001/ # scenario_type_instance_number
S00_sparse_fast_002/
S00_sparse_fast_003/
S01_sparse_slow_001/
...
S07_big_slow_012/ # 96 folders total (8 scenarios × 12)
Each folder contains:
input_component.csv # Component list (position, type, inspection time)
input_size.csv # PCB/FOV dimensions
input_parameter.csv # Equipment parameters (capture, recon, gantry speed)
input_step_region.csv # Step region coordinates
<solver>/ # Per-solver output subfolder (e.g. ky/, greedy/)
output_fov.csv # Solver output (FOV placement result)
result.html # Interactive viewer (3_view.py)
🎯 Scenario Type Matrix (8 Types × 12 = 96)
| ID | Layout Type | Recon Time | Components | Large Component % | Key Test Point |
|---|---|---|---|---|---|
| S00 | Sparse | 0.5s (fast) | 400–600 | 5–15% | Path optimization (baseline) |
| S01 | Sparse | 1.3s (slow) | 400–600 | 5–15% | Path + Recon Core utilization |
| S02 | Dense | 0.5s (fast) | 800–1200 | 5–15% | FOV packing efficiency |
| S03 | Dense | 1.3s (slow) | 800–1200 | 5–15% | Packing + Recon Core utilization |
| S04 | Grid | 0.5s (fast) | 1000–1500 | 0% | Scaling (uniform grid layout) |
| S05 | Grid | 1.3s (slow) | 1000–1500 | 0% | Scaling + Recon Core utilization |
| S06 | Big | 0.5s (fast) | 300–500 | 40–60% | Tiling logic (many large components) |
| S07 | Big | 1.3s (slow) | 300–500 | 40–60% | Tiling + Recon Core utilization |
💡
Recon time difference: When recon_time is 0.5s (fast), camera movement is the bottleneck. When it is 1.3s (slow), 3D reconstruction becomes the bottleneck. In slow-recon scenarios, strategies that utilize additional Recon Cores are critical.
⚠
Dataset notice: The 96 datasets (8 scenarios × 12) are provided for development and testing. You may also generate additional scenarios with
1_generate_data.py. The final contest evaluation will use a separate, hidden test set, which will be released on the GitHub repository after the contest ends.
2 Input Files
Four CSV files that the solver must read.
📋 input_component.csv — Component List
| Column | Type | Description |
|---|---|---|
(index) | int | Component index (0-based, row order) |
tl_x, tl_y | float | Top-left coordinates of the component bounding box (mm) |
br_x, br_y | float | Bottom-right coordinates of the component bounding box (mm) |
type | int | 2 = Fiducial 1 = Barcode 0 = Normal |
side | int | Side camera required (0: no, 1: yes) |
step | int | Step height (0: default, 1: elevated region) |
time | float | Per-component inspection time (seconds) |
,tl_x,tl_y,br_x,br_y,type,side,step,time
0,491.00,497.00,496.00,502.00,2,0,0,0.25 # Fiducial (type=2)
1,187.56,159.67,189.70,164.36,1,0,0,0.10 # Barcode (type=1)
2,102.88,191.31,104.01,196.56,1,0,0,0.059 # Barcode (type=1)
3,17.93,262.29,19.55,263.85,1,0,0,0.025 # Barcode (type=1)
4,396.32,18.40,398.48,20.57,0,0,0,0.015 # Normal component (type=0)
📐 input_size.csv — PCB/FOV Dimensions
| Column | Description | Value |
|---|---|---|
axis | Axis (x or y) | x, y |
pcb_size | PCB size (mm) | 128–512 (random per axis) |
fov_size | FOV size (mm) | 50–100 (random per axis) |
axis,pcb_size,fov_size
x,380,62 # pcb_size: 128–512 random per axis, fov_size: 50–100 random
y,280,55 # independent per axis
⚙ input_parameter.csv — Equipment Parameters
| Column | Description | Range |
|---|---|---|
capture_time | FOV capture time (seconds) | 0.5 (fixed) |
recon_time | 3D reconstruction time (seconds) | 0.5 (fast) or 1.3 (slow) |
side_capture_time | Side capture time (seconds) | 0.75 (fixed) |
max_core | Maximum parallel cores | 4–16 (random per dataset) |
v_x, v_y | Gantry max velocity (mm/s) | 200–1000 (random per dataset) |
a_x, a_y | Gantry acceleration (mm/s²) | 4900–9800 (random per dataset) |
capture_time,recon_time,side_capture_time,max_core,v_x,v_y,a_x,a_y
0.5,0.5,0.75,8,721,315,7900,9198 # fast scenario, max_core/v/a random per dataset
📏 input_step_region.csv — Step Regions
Defines regions on the PCB with different heights (step=1). If a component's bounding box overlaps with this region, that component has step=1.
| Column | Description |
|---|---|
(index) | Region index |
tl_x, tl_y | Top-left coordinates of the step region (mm) |
br_x, br_y | Bottom-right coordinates of the step region (mm) |
,tl_x,tl_y,br_x,br_y
0,353.89,214.80,486.21,382.28 # 1 step region (132mm × 167mm)
⚠
Step transition penalty: Transitioning between step=0 and step=1 FOVs incurs a +5 second penalty for Z-axis height adjustment. Visiting FOVs of the same step consecutively is more efficient.
3 Output File
The file that the solver must generate. After running
2_run.py --solver <name>, it is stored under simulation_data/<dataset>/<solver>/output_fov.csv.📤 output_fov.csv — FOV Placement Result
Row order = FOV visit order. The first row is the first FOV to be visited.
| Column | Type | Description |
|---|---|---|
(index) | int | FOV order (row order = visit order) |
x, y | float | FOV center coordinates (mm) |
comp_idx | JSON array | List of component indices contained in this FOV |
,x,y,comp_idx
0,475.10,487.00,"[280, 0, 306, 586, 118, 153, 169]" # 1st FOV: 7 components
1,188.63,162.01,"[139, 1, 550, 328, 180]" # 2nd FOV
2,408.85,30.58,"[4, 525, 377, 98]" # 3rd FOV
3,103.45,193.94,"[2]" # 4th FOV: 1 component
✅
Constraints: Every component must be included in at least one FOV (
cover_all_components), and the visit order must follow Fiducial → Barcode → Normal (fov_inspection_order).
4 PCB Coordinate System
The origin is at the top-left corner of the PCB (0, 0). The X-axis increases to the right, and the Y-axis increases downward.
5 Common Parameters
Fixed values shared across all scenarios.
📐 Dimensions & Speed
| Parameter | Value |
|---|---|
| PCB size | 128~512 × 128~512 mm (random per axis) |
| FOV size | 50–100 × 50–100 mm (random per axis) |
| Capture time (capture_time) | 0.5 s |
| Side capture time | 0.75 s |
| Gantry velocity (v_x, v_y) | 200–1000 mm/s (random per dataset) |
| Gantry acceleration (a_x, a_y) | 4900–9800 mm/s² (random per dataset) |
🔧 Per-Scenario Variables
| Parameter | Range |
|---|---|
| max_core | 4–16 (random per dataset) |
| recon_time | 0.5 s (fast) / 1.3 s (slow) |
| Component count | 300–1500 (varies by type) |
| Large component ratio | 0–60% (varies by type) |
| Side component ratio | ~20% |
| Step regions | 1–2 |
This dataset is licensed under CC BY-NC-SA 4.0. See Problem Description §7 for details.