finished item finder

This commit is contained in:
Boki 2026-02-10 19:10:59 -05:00
parent 1246884be9
commit 930e00c9cc
7 changed files with 450 additions and 37 deletions

1
.gitignore vendored
View file

@ -7,6 +7,7 @@ browser-data/
debug-screenshots/
eng.traineddata
.claude/
nul
# OcrDaemon build output
tools/OcrDaemon/bin/

View file

@ -4,6 +4,7 @@ import { WebSocketServer, WebSocket } from 'ws';
import path from 'path';
import { fileURLToPath } from 'url';
import { logger } from '../util/logger.js';
import { sleep } from '../util/sleep.js';
import type { BotController } from './BotController.js';
import type { ScreenReader } from '../game/ScreenReader.js';
import { GRID_LAYOUTS } from '../game/GridReader.js';
@ -203,18 +204,21 @@ export class DashboardServer {
// Grid scan with calibrated positions
this.app.post('/api/debug/grid-scan', async (req, res) => {
if (!this.debug) { res.status(503).json({ error: 'Debug not available' }); return; }
const { layout: layoutName } = req.body as { layout: string };
const { layout: layoutName, targetRow, targetCol } = req.body as { layout: string; targetRow?: number; targetCol?: number };
if (!GRID_LAYOUTS[layoutName]) { res.status(400).json({ error: `Unknown grid layout: ${layoutName}` }); return; }
try {
const result = await this.debug.screenReader.grid.scan(layoutName);
const result = await this.debug.screenReader.grid.scan(layoutName, undefined, targetRow, targetCol);
const imageBuffer = await this.debug.screenReader.captureRegion(result.layout.region);
const imageBase64 = imageBuffer.toString('base64');
const r = result.layout.region;
const matchInfo = result.matches ? `, ${result.matches.length} matches` : '';
this.broadcastLog('info',
`Grid scan (${layoutName}): ${result.layout.cols}x${result.layout.rows} at (${r.x},${r.y}) ${r.width}x${r.height}${result.occupied.length} occupied cells`);
`Grid scan (${layoutName}): ${result.layout.cols}x${result.layout.rows} at (${r.x},${r.y}) ${r.width}x${r.height}${result.occupied.length} occupied cells${matchInfo}`);
res.json({
ok: true,
occupied: result.occupied,
items: result.items,
matches: result.matches,
cols: result.layout.cols,
rows: result.layout.rows,
image: imageBase64,
@ -247,6 +251,51 @@ export class DashboardServer {
}
});
// Test: scan grid, find matches for target cell, hover over each for 1s
this.app.post('/api/debug/test-match-hover', async (req, res) => {
if (!this.debug) { res.status(503).json({ error: 'Debug not available' }); return; }
const { layout: layoutName, targetRow, targetCol } = req.body as { layout: string; targetRow: number; targetCol: number };
if (!GRID_LAYOUTS[layoutName]) { res.status(400).json({ error: `Unknown layout: ${layoutName}` }); return; }
if (targetRow == null || targetCol == null) { res.status(400).json({ error: 'Missing targetRow/targetCol' }); return; }
try {
// Scan with match target
this.broadcastLog('info', `Scanning ${layoutName} with target (${targetRow},${targetCol})...`);
const result = await this.debug.screenReader.grid.scan(layoutName, undefined, targetRow, targetCol);
const matches = result.matches ?? [];
const items = result.items ?? [];
// Find the item dimensions at target cell
const targetItem = items.find(i => targetRow >= i.row && targetRow < i.row + i.h && targetCol >= i.col && targetCol < i.col + i.w);
const itemSize = targetItem ? `${targetItem.w}x${targetItem.h}` : '1x1';
this.broadcastLog('info', `Target (${targetRow},${targetCol}) is ${itemSize}, found ${matches.length} matches`);
// Build list of cells to hover: target first, then matches
const hoverCells = [
{ row: targetRow, col: targetCol, label: 'TARGET' },
...matches.map(m => ({ row: m.row, col: m.col, label: `MATCH ${(m.similarity * 100).toFixed(0)}%` })),
];
// Focus game and hover each cell
await this.debug.gameController.focusGame();
for (const cell of hoverCells) {
const center = result.layout.region;
const cellW = center.width / result.layout.cols;
const cellH = center.height / result.layout.rows;
const x = Math.round(center.x + cell.col * cellW + cellW / 2);
const y = Math.round(center.y + cell.row * cellH + cellH / 2);
this.broadcastLog('info', `Hovering ${cell.label} (${cell.row},${cell.col}) at (${x},${y})...`);
await this.debug.gameController.moveMouseTo(x, y);
await sleep(1000);
}
this.broadcastLog('info', `Done — hovered ${hoverCells.length} cells`);
res.json({ ok: true, itemSize, matchCount: matches.length, hoveredCount: hoverCells.length });
} catch (err) {
logger.error({ err }, 'Debug test-match-hover failed');
res.status(500).json({ error: 'Test match hover failed' });
}
});
this.server = http.createServer(this.app);
this.wss = new WebSocketServer({ server: this.server });

View file

@ -397,6 +397,19 @@
<button onclick="debugGridScan('shop')">Scan Shop</button>
<button onclick="debugGridScan('vendor')">Scan Vendor</button>
</div>
<div class="debug-row">
<select id="matchLayout" style="padding:6px 10px;background:#0d1117;border:1px solid #30363d;border-radius:6px;color:#e6edf3;font-size:13px">
<option value="shop">Shop</option>
<option value="seller">Seller</option>
<option value="stash12">Stash 12×12</option>
<option value="stash12_folder">Stash 12×12 (folder)</option>
<option value="inventory">Inventory</option>
<option value="vendor">Vendor</option>
</select>
<input type="number" id="matchRow" placeholder="Row" value="8" style="width:60px" />
<input type="number" id="matchCol" placeholder="Col" value="0" style="width:60px" />
<button class="primary" onclick="debugTestMatchHover()">Match & Hover</button>
</div>
<div class="debug-row">
<input type="text" id="debugTextInput" placeholder="Text to find (e.g. Stash, Ange)" />
<button onclick="debugFindText()">Find</button>
@ -645,33 +658,81 @@
}
// Debug functions
async function debugGridScan(layout) {
showDebugResult(`Scanning ${layout}...`);
let lastGridLayout = null;
async function debugGridScan(layout, targetRow, targetCol) {
lastGridLayout = layout;
const hasTarget = targetRow != null && targetCol != null;
showDebugResult(hasTarget ? `Matching (${targetRow},${targetCol}) in ${layout}...` : `Scanning ${layout}...`);
const body = { layout };
if (hasTarget) { body.targetRow = targetRow; body.targetCol = targetCol; }
const res = await fetch('/api/debug/grid-scan', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ layout }),
body: JSON.stringify(body),
});
const data = await res.json();
if (!data.ok) { showDebugResult(`Error: ${data.error}`); return; }
const el = document.getElementById('debugResult');
const count = data.occupied.length;
const items = data.items || [];
const matches = data.matches || [];
const r = data.region;
let html = `<b>${layout}</b> ${data.cols}x${data.rows}`;
html += ` — ${count} occupied cell(s)`;
html += ` — ${count} occupied, ${items.length} items`;
if (matches.length > 0) html += `, <span style="color:#f0883e;font-weight:bold">${matches.length} matches</span>`;
if (r) html += `<br><span style="color:#484f58">Region: (${r.x}, ${r.y}) ${r.width}x${r.height}</span>`;
if (count > 0) {
html += `<br>` + data.occupied.map(c => `(${c.row},${c.col})`).join(' ');
if (items.length > 0) {
const sizes = {};
items.forEach(i => { const k = i.w + 'x' + i.h; sizes[k] = (sizes[k]||0) + 1; });
html += `<br><span style="color:#58a6ff">` + Object.entries(sizes).map(([k,v]) => `${v}x ${k}`).join(', ') + `</span>`;
}
if (hasTarget) {
html += `<br><span style="color:#f0883e">Target: (${targetRow},${targetCol})`;
if (matches.length > 0) html += ` → ${matches.map(m => `(${m.row},${m.col}) ${(m.similarity*100).toFixed(0)}%`).join(', ')}`;
html += `</span>`;
} else {
html += `<br><span style="color:#484f58">Click a cell to find matching items</span>`;
}
html += '<div class="grid-debug">';
if (data.image) {
html += `<img src="data:image/png;base64,${data.image}" alt="Grid capture" />`;
}
// Build match set for highlighting
const matchSet = new Set(matches.map(m => m.row + ',' + m.col));
const targetKey = hasTarget ? targetRow + ',' + targetCol : null;
// Build item map: cell → item info
const itemMap = {};
const colors = ['#238636','#1f6feb','#8957e5','#da3633','#d29922','#3fb950','#388bfd','#a371f7','#f85149','#e3b341'];
items.forEach((item, idx) => {
const color = colors[idx % colors.length];
for (let dr = 0; dr < item.h; dr++)
for (let dc = 0; dc < item.w; dc++)
itemMap[(item.row+dr)+','+(item.col+dc)] = { item, color, isOrigin: dr===0 && dc===0 };
});
html += `<div class="grid-view" style="grid-template-columns:repeat(${data.cols},12px)">`;
const set = new Set(data.occupied.map(c => c.row + ',' + c.col));
for (let r = 0; r < data.rows; r++) {
for (let c = 0; c < data.cols; c++) {
html += `<div class="grid-cell${set.has(r+','+c) ? ' occupied' : ''}" title="(${r},${c})"></div>`;
const key = r+','+c;
const isTarget = key === targetKey;
const isMatch = matchSet.has(key);
const info = itemMap[key];
let bg;
if (isTarget) bg = '#f0883e';
else if (isMatch) bg = '#d29922';
else if (info) bg = info.color;
else if (set.has(key)) bg = '#238636';
else bg = '';
const outline = (isTarget || isMatch) ? 'outline:2px solid #f0883e;z-index:1;' : '';
const cursor = set.has(key) ? 'cursor:pointer;' : '';
const bgStyle = bg ? `background:${bg};` : '';
const style = (bgStyle || outline || cursor) ? ` style="${bgStyle}${outline}${cursor}"` : '';
let title = info ? `(${r},${c}) ${info.item.w}x${info.item.h}` : `(${r},${c})`;
if (isTarget) title += ' [TARGET]';
if (isMatch) { const m = matches.find(m => m.row===r && m.col===c); title += ` [MATCH ${(m.similarity*100).toFixed(0)}%]`; }
const onclick = set.has(key) ? ` onclick="debugGridScan('${layout}',${r},${c})"` : '';
html += `<div class="grid-cell${set.has(key) ? ' occupied' : ''}"${style}${onclick} title="${title}"></div>`;
}
}
html += '</div></div>';
@ -764,6 +825,22 @@
showDebugResult(data.ok ? `Clicked at (${x}, ${y})` : `Error: ${data.error}`);
}
async function debugTestMatchHover() {
const layout = document.getElementById('matchLayout').value;
const targetRow = parseInt(document.getElementById('matchRow').value);
const targetCol = parseInt(document.getElementById('matchCol').value);
if (isNaN(targetRow) || isNaN(targetCol)) { showDebugResult('Invalid row/col'); return; }
showDebugResult(`Scanning ${layout} and matching (${targetRow},${targetCol})...`);
const res = await fetch('/api/debug/test-match-hover', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ layout, targetRow, targetCol }),
});
const data = await res.json();
if (!data.ok) { showDebugResult(`Error: ${data.error}`); return; }
showDebugResult(`Done: ${data.itemSize} item, ${data.matchCount} matches, hovered ${data.hoveredCount} cells`);
}
function showDebugResult(text) {
document.getElementById('debugResult').textContent = text;
}

View file

@ -87,6 +87,10 @@ export class GameController {
await this.inputSender.ctrlRightClick(x, y);
}
async moveMouseTo(x: number, y: number): Promise<void> {
await this.inputSender.moveMouse(x, y);
}
async leftClickAt(x: number, y: number): Promise<void> {
await this.inputSender.leftClick(x, y);
}

View file

@ -1,5 +1,5 @@
import { logger } from '../util/logger.js';
import type { OcrDaemon } from './OcrDaemon.js';
import type { OcrDaemon, GridItem, GridMatch } from './OcrDaemon.js';
import type { Region } from '../types.js';
// ── Grid type definitions ───────────────────────────────────────────────────
@ -20,6 +20,8 @@ export interface CellCoord {
export interface ScanResult {
layout: GridLayout;
occupied: CellCoord[];
items: GridItem[];
matches?: GridMatch[];
}
// ── Calibrated grid layouts (2560×1440) ─────────────────────────────────────
@ -89,20 +91,20 @@ export class GridReader {
/**
* Scan a named grid layout for occupied cells.
*/
async scan(layoutName: string, threshold?: number): Promise<ScanResult> {
async scan(layoutName: string, threshold?: number, targetRow?: number, targetCol?: number): Promise<ScanResult> {
const layout = GRID_LAYOUTS[layoutName];
if (!layout) throw new Error(`Unknown grid layout: ${layoutName}`);
const t = performance.now();
const occupied = await this.getOccupiedCells(layout, threshold);
const { occupied, items, matches } = await this.getOccupiedCells(layout, threshold, targetRow, targetCol);
const ms = (performance.now() - t).toFixed(0);
logger.info(
{ layoutName, cols: layout.cols, rows: layout.rows, occupied: occupied.length, ms },
{ layoutName, cols: layout.cols, rows: layout.rows, occupied: occupied.length, items: items.length, matches: matches?.length, ms },
'Grid scan complete',
);
return { layout, occupied };
return { layout, occupied, items, matches };
}
/** Get the screen-space center of a grid cell */
@ -115,20 +117,22 @@ export class GridReader {
};
}
/** Scan the grid and return which cells are occupied */
async getOccupiedCells(layout: GridLayout, threshold?: number): Promise<CellCoord[]> {
/** Scan the grid and return which cells are occupied and detected items */
async getOccupiedCells(layout: GridLayout, threshold?: number, targetRow?: number, targetCol?: number): Promise<{ occupied: CellCoord[]; items: GridItem[]; matches?: GridMatch[] }> {
const t = performance.now();
const cells = await this.daemon.gridScan(
const result = await this.daemon.gridScan(
layout.region,
layout.cols,
layout.rows,
threshold,
targetRow,
targetCol,
);
const occupied: CellCoord[] = [];
for (let row = 0; row < cells.length; row++) {
for (let col = 0; col < cells[row].length; col++) {
if (cells[row][col]) {
for (let row = 0; row < result.cells.length; row++) {
for (let col = 0; col < result.cells[row].length; col++) {
if (result.cells[row][col]) {
const center = this.getCellCenter(layout, row, col);
occupied.push({ row, col, x: center.x, y: center.y });
}
@ -137,10 +141,10 @@ export class GridReader {
const ms = (performance.now() - t).toFixed(0);
logger.info(
{ layout: `${layout.cols}x${layout.rows}`, occupied: occupied.length, ms },
{ layout: `${layout.cols}x${layout.rows}`, occupied: occupied.length, items: result.items.length, matches: result.matches?.length, ms },
'Grid scan complete',
);
return occupied;
return { occupied, items: result.items, matches: result.matches };
}
/** Get all cell centers in the grid */

View file

@ -24,6 +24,25 @@ export interface OcrResponse {
lines: OcrLine[];
}
export interface GridItem {
row: number;
col: number;
w: number;
h: number;
}
export interface GridMatch {
row: number;
col: number;
similarity: number;
}
export interface GridScanResult {
cells: boolean[][];
items: GridItem[];
matches?: GridMatch[];
}
export interface DetectGridResult {
detected: boolean;
region?: Region;
@ -51,6 +70,8 @@ interface DaemonResponse {
lines?: OcrLine[];
image?: string;
cells?: boolean[][];
items?: GridItem[];
matches?: GridMatch[];
detected?: boolean;
region?: Region;
cols?: number;
@ -106,11 +127,13 @@ export class OcrDaemon {
return Buffer.from(resp.image!, 'base64');
}
async gridScan(region: Region, cols: number, rows: number, threshold?: number): Promise<boolean[][]> {
async gridScan(region: Region, cols: number, rows: number, threshold?: number, targetRow?: number, targetCol?: number): Promise<GridScanResult> {
const req: DaemonRequest = { cmd: 'grid', region, cols, rows };
if (threshold) req.threshold = threshold;
if (targetRow != null && targetRow >= 0) (req as any).targetRow = targetRow;
if (targetCol != null && targetCol >= 0) (req as any).targetCol = targetCol;
const resp = await this.sendWithRetry(req, REQUEST_TIMEOUT);
return resp.cells ?? [];
return { cells: resp.cells ?? [], items: resp.items ?? [], matches: resp.matches ?? undefined };
}
async detectGrid(region: Region, minCellSize?: number, maxCellSize?: number): Promise<DetectGridResult> {

View file

@ -129,10 +129,13 @@ void HandleCapture(Request req)
}
// Pre-loaded empty cell templates (loaded lazily on first grid scan)
// Stored as both grayscale (for occupied detection) and ARGB (for item border detection)
byte[]? emptyTemplate70Gray = null;
int emptyTemplate70W = 0, emptyTemplate70H = 0;
byte[]? emptyTemplate70Argb = null;
int emptyTemplate70W = 0, emptyTemplate70H = 0, emptyTemplate70Stride = 0;
byte[]? emptyTemplate35Gray = null;
int emptyTemplate35W = 0, emptyTemplate35H = 0;
byte[]? emptyTemplate35Argb = null;
int emptyTemplate35W = 0, emptyTemplate35H = 0, emptyTemplate35Stride = 0;
void LoadTemplatesIfNeeded()
{
@ -150,23 +153,23 @@ void LoadTemplatesIfNeeded()
using var bmp = new Bitmap(t70Path);
emptyTemplate70W = bmp.Width;
emptyTemplate70H = bmp.Height;
emptyTemplate70Gray = BitmapToGray(bmp);
(emptyTemplate70Gray, emptyTemplate70Argb, emptyTemplate70Stride) = BitmapToGrayAndArgb(bmp);
}
if (System.IO.File.Exists(t35Path))
{
using var bmp = new Bitmap(t35Path);
emptyTemplate35W = bmp.Width;
emptyTemplate35H = bmp.Height;
emptyTemplate35Gray = BitmapToGray(bmp);
(emptyTemplate35Gray, emptyTemplate35Argb, emptyTemplate35Stride) = BitmapToGrayAndArgb(bmp);
}
}
byte[] BitmapToGray(Bitmap bmp)
(byte[] gray, byte[] argb, int stride) BitmapToGrayAndArgb(Bitmap bmp)
{
int w = bmp.Width, h = bmp.Height;
var data = bmp.LockBits(new Rectangle(0, 0, w, h), ImageLockMode.ReadOnly, PixelFormat.Format32bppArgb);
byte[] pixels = new byte[data.Stride * h];
Marshal.Copy(data.Scan0, pixels, 0, pixels.Length);
byte[] argb = new byte[data.Stride * h];
Marshal.Copy(data.Scan0, argb, 0, argb.Length);
bmp.UnlockBits(data);
int stride = data.Stride;
@ -175,8 +178,14 @@ byte[] BitmapToGray(Bitmap bmp)
for (int x = 0; x < w; x++)
{
int i = y * stride + x * 4;
gray[y * w + x] = (byte)((pixels[i] + pixels[i + 1] + pixels[i + 2]) / 3);
gray[y * w + x] = (byte)((argb[i] + argb[i + 1] + argb[i + 2]) / 3);
}
return (gray, argb, stride);
}
byte[] BitmapToGray(Bitmap bmp)
{
var (gray, _, _) = BitmapToGrayAndArgb(bmp);
return gray;
}
@ -199,18 +208,23 @@ void HandleGrid(Request req)
// Pick the right empty template based on cell size
int nominalCell = (int)Math.Round(cellW);
byte[]? templateGray;
int templateW, templateH;
byte[]? templateArgb;
int templateW, templateH, templateStride;
if (nominalCell <= 40 && emptyTemplate35Gray != null)
{
templateGray = emptyTemplate35Gray;
templateArgb = emptyTemplate35Argb!;
templateW = emptyTemplate35W;
templateH = emptyTemplate35H;
templateStride = emptyTemplate35Stride;
}
else if (emptyTemplate70Gray != null)
{
templateGray = emptyTemplate70Gray;
templateArgb = emptyTemplate70Argb!;
templateW = emptyTemplate70W;
templateH = emptyTemplate70H;
templateStride = emptyTemplate70Stride;
}
else
{
@ -218,8 +232,8 @@ void HandleGrid(Request req)
return;
}
// Convert captured bitmap to grayscale
byte[] captureGray = BitmapToGray(bitmap);
// Convert captured bitmap to grayscale + keep ARGB for border color comparison
var (captureGray, captureArgb, captureStride) = BitmapToGrayAndArgb(bitmap);
int captureW = bitmap.Width;
// Border to skip (outer pixels may differ between cells)
@ -277,7 +291,209 @@ void HandleGrid(Request req)
if (debug) Console.Error.WriteLine($" Row {row,2}: {string.Join(" ", debugDiffs)}");
}
WriteResponse(new GridResponse { Cells = cells });
// ── Item detection: compare border pixels to empty template (grayscale) ──
// Items have a colored tint behind them that shows through grid lines.
// Compare each cell's border strip against the template's border pixels.
// If they differ → item tint present → cells belong to same item.
int[] parent = new int[rows * cols];
for (int i = 0; i < parent.Length; i++) parent[i] = i;
int Find(int x) { while (parent[x] != x) { parent[x] = parent[parent[x]]; x = parent[x]; } return x; }
void Union(int a, int b) { parent[Find(a)] = Find(b); }
int stripWidth = Math.Max(2, border / 2);
int stripInset = (int)(cellW * 0.15);
double borderDiffThresh = 15.0;
for (int row = 0; row < rows; row++)
{
for (int col = 0; col < cols; col++)
{
if (!cells[row][col]) continue;
int cx0 = (int)(col * cellW);
int cy0 = (int)(row * cellH);
// Check right neighbor
if (col + 1 < cols && cells[row][col + 1])
{
long diffSum = 0; int cnt = 0;
int xStart = (int)((col + 1) * cellW) - stripWidth;
int yFrom = cy0 + stripInset;
int yTo = (int)((row + 1) * cellH) - stripInset;
for (int sy = yFrom; sy < yTo; sy += 2)
{
int tmplY = sy - cy0;
for (int sx = xStart; sx < xStart + stripWidth * 2; sx++)
{
if (sx < 0 || sx >= captureW) continue;
int tmplX = sx - cx0;
if (tmplX < 0 || tmplX >= templateW) continue;
diffSum += Math.Abs(captureGray[sy * captureW + sx] - templateGray[tmplY * templateW + tmplX]);
cnt++;
}
}
double meanDiff = cnt > 0 ? (double)diffSum / cnt : 0;
if (debug) Console.Error.WriteLine($" H ({row},{col})->({row},{col+1}): {meanDiff:F1}{(meanDiff > borderDiffThresh ? " SAME" : "")}");
if (meanDiff > borderDiffThresh)
Union(row * cols + col, row * cols + col + 1);
}
// Check bottom neighbor
if (row + 1 < rows && cells[row + 1][col])
{
long diffSum = 0; int cnt = 0;
int yStart = (int)((row + 1) * cellH) - stripWidth;
int xFrom = cx0 + stripInset;
int xTo = (int)((col + 1) * cellW) - stripInset;
for (int sx = xFrom; sx < xTo; sx += 2)
{
int tmplX = sx - cx0;
for (int sy = yStart; sy < yStart + stripWidth * 2; sy++)
{
if (sy < 0 || sy >= bitmap.Height) continue;
int tmplY = sy - cy0;
if (tmplY < 0 || tmplY >= templateH) continue;
diffSum += Math.Abs(captureGray[sy * captureW + sx] - templateGray[tmplY * templateW + tmplX]);
cnt++;
}
}
double meanDiff = cnt > 0 ? (double)diffSum / cnt : 0;
if (debug) Console.Error.WriteLine($" V ({row},{col})->({row+1},{col}): {meanDiff:F1}{(meanDiff > borderDiffThresh ? " SAME" : "")}");
if (meanDiff > borderDiffThresh)
Union(row * cols + col, (row + 1) * cols + col);
}
}
}
// Extract items from union-find groups
var groups = new Dictionary<int, List<(int row, int col)>>();
for (int row = 0; row < rows; row++)
for (int col = 0; col < cols; col++)
if (cells[row][col])
{
int root = Find(row * cols + col);
if (!groups.ContainsKey(root)) groups[root] = [];
groups[root].Add((row, col));
}
var items = new List<GridItem>();
foreach (var group in groups.Values)
{
int minR = group.Min(c => c.row);
int maxR = group.Max(c => c.row);
int minC = group.Min(c => c.col);
int maxC = group.Max(c => c.col);
items.Add(new GridItem { Row = minR, Col = minC, W = maxC - minC + 1, H = maxR - minR + 1 });
}
if (debug)
{
Console.Error.WriteLine($" Items found: {items.Count}");
foreach (var item in items)
Console.Error.WriteLine($" ({item.Row},{item.Col}) {item.W}x{item.H}");
}
// ── Visual matching: find cells similar to target ──
List<GridMatch>? matches = null;
if (req.TargetRow >= 0 && req.TargetCol >= 0 &&
req.TargetRow < rows && req.TargetCol < cols &&
cells[req.TargetRow][req.TargetCol])
{
matches = FindMatchingCells(
captureGray, captureW, bitmap.Height,
cells, rows, cols, cellW, cellH, border,
req.TargetRow, req.TargetCol, debug);
}
WriteResponse(new GridResponse { Cells = cells, Items = items, Matches = matches });
}
/// Find all occupied cells visually similar to the target cell using full-resolution NCC.
/// Full resolution gives better discrimination — sockets are a small fraction of total pixels.
List<GridMatch> FindMatchingCells(
byte[] gray, int imgW, int imgH,
List<List<bool>> cells, int rows, int cols,
float cellW, float cellH, int border,
int targetRow, int targetCol, bool debug)
{
int innerW = (int)cellW - border * 2;
int innerH = (int)cellH - border * 2;
if (innerW <= 4 || innerH <= 4) return [];
int tCx0 = (int)(targetCol * cellW) + border;
int tCy0 = (int)(targetRow * cellH) + border;
int tInnerW = Math.Min(innerW, imgW - tCx0);
int tInnerH = Math.Min(innerH, imgH - tCy0);
if (tInnerW < innerW || tInnerH < innerH) return [];
int n = innerW * innerH;
// Pre-compute target cell pixels and stats
double[] targetPixels = new double[n];
double tMean = 0;
for (int py = 0; py < innerH; py++)
for (int px = 0; px < innerW; px++)
{
double v = gray[(tCy0 + py) * imgW + (tCx0 + px)];
targetPixels[py * innerW + px] = v;
tMean += v;
}
tMean /= n;
double tStd = 0;
for (int i = 0; i < n; i++)
tStd += (targetPixels[i] - tMean) * (targetPixels[i] - tMean);
tStd = Math.Sqrt(tStd / n);
if (debug) Console.Error.WriteLine($" Match target ({targetRow},{targetCol}): {innerW}x{innerH} ({n}px), mean={tMean:F1}, std={tStd:F1}");
if (tStd < 3.0) return [];
double matchThreshold = 0.70;
var matches = new List<GridMatch>();
for (int row = 0; row < rows; row++)
{
for (int col = 0; col < cols; col++)
{
if (!cells[row][col]) continue;
if (row == targetRow && col == targetCol) continue;
int cx0 = (int)(col * cellW) + border;
int cy0 = (int)(row * cellH) + border;
int cInnerW = Math.Min(innerW, imgW - cx0);
int cInnerH = Math.Min(innerH, imgH - cy0);
if (cInnerW < innerW || cInnerH < innerH) continue;
// Compute NCC at full resolution
double cMean = 0;
for (int py = 0; py < innerH; py++)
for (int px = 0; px < innerW; px++)
cMean += gray[(cy0 + py) * imgW + (cx0 + px)];
cMean /= n;
double cStd = 0, cross = 0;
for (int py = 0; py < innerH; py++)
for (int px = 0; px < innerW; px++)
{
double cv = gray[(cy0 + py) * imgW + (cx0 + px)] - cMean;
double tv = targetPixels[py * innerW + px] - tMean;
cStd += cv * cv;
cross += tv * cv;
}
cStd = Math.Sqrt(cStd / n);
double ncc = (tStd > 0 && cStd > 0) ? cross / (n * tStd * cStd) : 0;
if (debug && ncc > 0.5)
Console.Error.WriteLine($" ({row},{col}): NCC={ncc:F3}");
if (ncc >= matchThreshold)
matches.Add(new GridMatch { Row = row, Col = col, Similarity = Math.Round(ncc, 3) });
}
}
if (debug) Console.Error.WriteLine($" Matches for ({targetRow},{targetCol}): {matches.Count}");
return matches;
}
void HandleDetectGrid(Request req)
@ -788,6 +1004,12 @@ class Request
[JsonPropertyName("debug")]
public bool Debug { get; set; }
[JsonPropertyName("targetRow")]
public int TargetRow { get; set; } = -1;
[JsonPropertyName("targetCol")]
public int TargetCol { get; set; } = -1;
}
class RegionRect
@ -884,6 +1106,39 @@ class GridResponse
[JsonPropertyName("cells")]
public List<List<bool>> Cells { get; set; } = [];
[JsonPropertyName("items")]
public List<GridItem>? Items { get; set; }
[JsonPropertyName("matches")]
public List<GridMatch>? Matches { get; set; }
}
class GridItem
{
[JsonPropertyName("row")]
public int Row { get; set; }
[JsonPropertyName("col")]
public int Col { get; set; }
[JsonPropertyName("w")]
public int W { get; set; }
[JsonPropertyName("h")]
public int H { get; set; }
}
class GridMatch
{
[JsonPropertyName("row")]
public int Row { get; set; }
[JsonPropertyName("col")]
public int Col { get; set; }
[JsonPropertyName("similarity")]
public double Similarity { get; set; }
}
class DetectGridResponse