Files
ui/src/features/annotations/CanvasEditor.tsx
T
Oleksandr Bezdieniezhnykh 085d7bf17e Merge API remote base URL config into dev
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-06-24 18:19:02 +03:00

520 lines
18 KiB
TypeScript

import { useRef, useEffect, useState, useCallback, forwardRef, useImperativeHandle } from 'react'
import { endpoints, authenticatedApiUrl } from '../../api'
import { MediaType } from '../../types'
import type { Media, AnnotationListItem, Detection, Affiliation, CombatReadiness } from '../../types'
import { getClassColor, getClassNameFallback, hexToRgba } from '../../class-colors'
import { parseAnnotationTime } from './time'
interface Props {
media: Media
annotation: AnnotationListItem | null
detections: Detection[]
onDetectionsChange: (dets: Detection[]) => void
selectedClassNum: number
currentTime: number
annotations: AnnotationListItem[]
onZoomChange?: (zoom: number) => void
onCursorChange?: (nx: number, ny: number) => void
}
export interface CanvasEditorHandle {
deleteSelected: () => void
deleteAll: () => void
hasSelection: () => boolean
}
interface DragState {
type: 'draw' | 'move' | 'resize'
startX: number
startY: number
detectionIndex?: number
handle?: string
}
interface LabelChip {
leftPct: number
topPct: number
color: string
name: string
conf: number
combatReady: boolean
}
const HANDLE_SIZE = 6
const MIN_BOX_SIZE = 12
const HOSTILE_HEXES = new Set(['#FF0000', '#FFFF00', '#FF00FF', '#800000', '#808000', '#800080'])
const FRIENDLY_HEXES = new Set(['#00FF00', '#0000FF', '#00FFFF', '#008000', '#000080', '#008080'])
function affiliationIcon(hex: string) {
const up = hex.toUpperCase()
if (HOSTILE_HEXES.has(up)) {
return (
<svg width="11" height="11" viewBox="0 0 11 11" aria-hidden="true">
<polygon points="5.5,0.7 10.3,5.5 5.5,10.3 0.7,5.5" fill="#FF0000" stroke="#0A0D10" strokeWidth="1"/>
</svg>
)
}
if (FRIENDLY_HEXES.has(up)) {
return (
<svg width="11" height="9" viewBox="0 0 11 9" aria-hidden="true">
<rect x="0.5" y="0.5" width="10" height="8" fill="#87CEEB" stroke="#0A0D10" strokeWidth="1"/>
</svg>
)
}
return (
<svg width="10" height="10" viewBox="0 0 10 10" aria-hidden="true">
<circle cx="5" cy="5" r="3.5" fill="none" stroke="currentColor" strokeWidth="1.2"/>
</svg>
)
}
const CanvasEditor = forwardRef<CanvasEditorHandle, Props>(function CanvasEditor(
{ media, annotation, detections, onDetectionsChange, selectedClassNum, currentTime, annotations, onZoomChange, onCursorChange },
ref,
) {
const canvasRef = useRef<HTMLCanvasElement>(null)
const containerRef = useRef<HTMLDivElement>(null)
const imgRef = useRef<HTMLImageElement | null>(null)
const cursorRafRef = useRef<number | null>(null)
const cursorLatestRef = useRef<{ x: number; y: number } | null>(null)
const [zoom, setZoom] = useState(1)
const [pan, setPan] = useState({ x: 0, y: 0 })
const [selected, setSelected] = useState<Set<number>>(new Set())
const [dragState, setDragState] = useState<DragState | null>(null)
const [drawRect, setDrawRect] = useState<{ x: number; y: number; w: number; h: number } | null>(null)
const [imgSize, setImgSize] = useState({ w: 0, h: 0 })
const [labelChips, setLabelChips] = useState<LabelChip[]>([])
useImperativeHandle(ref, () => ({
deleteSelected() {
if (selected.size === 0) return
onDetectionsChange(detections.filter((_, i) => !selected.has(i)))
setSelected(new Set())
},
deleteAll() {
onDetectionsChange([])
setSelected(new Set())
},
hasSelection() {
return selected.size > 0
},
}), [selected, detections, onDetectionsChange])
const isVideo = media.mediaType === MediaType.Video
const loadImage = useCallback(() => {
if (isVideo) {
imgRef.current = null
return
}
const img = new Image()
img.crossOrigin = 'anonymous'
const isLocalPath = media.path.startsWith('blob:') || media.path.startsWith('data:')
if (annotation && !isLocalPath) {
img.src = authenticatedApiUrl(endpoints.annotations.annotationImage(annotation.id))
} else if (isLocalPath) {
img.src = media.path
} else {
img.src = authenticatedApiUrl(endpoints.annotations.mediaFile(media.id))
}
img.onload = () => {
imgRef.current = img
const w = img.naturalWidth
const h = img.naturalHeight
setImgSize({ w, h })
const c = containerRef.current
if (c && w && h) {
const fit = Math.min(c.clientWidth / w, c.clientHeight / h)
const clamped = Math.max(0.05, Math.min(10, fit))
setZoom(clamped)
setPan({
x: (c.clientWidth - w * clamped) / 2,
y: (c.clientHeight - h * clamped) / 2,
})
}
}
}, [media, annotation, isVideo])
useEffect(() => { loadImage() }, [loadImage])
useEffect(() => {
if (!isVideo || !containerRef.current) return
const update = () => {
const c = containerRef.current
if (c) setImgSize({ w: c.clientWidth, h: c.clientHeight })
}
update()
const ro = new ResizeObserver(update)
ro.observe(containerRef.current)
return () => ro.disconnect()
}, [isVideo])
useEffect(() => { onZoomChange?.(zoom) }, [zoom, onZoomChange])
// Cancel any pending cursor RAF on unmount so the callback can't fire after.
useEffect(() => () => {
if (cursorRafRef.current != null) {
cancelAnimationFrame(cursorRafRef.current)
cursorRafRef.current = null
}
}, [])
const fromCanvas = useCallback((cx: number, cy: number) => ({
x: Math.max(0, Math.min(1, (cx - pan.x) / (imgSize.w * zoom))),
y: Math.max(0, Math.min(1, (cy - pan.y) / (imgSize.h * zoom))),
}), [imgSize, zoom, pan])
const getTimeWindowDetections = useCallback((): Detection[] => {
if (media.mediaType !== MediaType.Video) return []
if (annotation) return []
const timeTicks = currentTime * 10_000_000
return annotations
.filter(a => {
const sec = parseAnnotationTime(a.time)
if (sec == null) return false
return Math.abs(sec * 10_000_000 - timeTicks) < 2_000_000
})
.flatMap(a => a.detections)
}, [media.mediaType, annotation, annotations, currentTime])
const getHandles = (x: number, y: number, w: number, h: number) => [
{ x, y, cursor: 'nw-resize', name: 'tl' },
{ x: x + w / 2, y, cursor: 'n-resize', name: 'tc' },
{ x: x + w, y, cursor: 'ne-resize', name: 'tr' },
{ x: x + w, y: y + h / 2, cursor: 'e-resize', name: 'mr' },
{ x: x + w, y: y + h, cursor: 'se-resize', name: 'br' },
{ x: x + w / 2, y: y + h, cursor: 's-resize', name: 'bc' },
{ x, y: y + h, cursor: 'sw-resize', name: 'bl' },
{ x, y: y + h / 2, cursor: 'w-resize', name: 'ml' },
]
const draw = useCallback(() => {
const canvas = canvasRef.current
const ctx = canvas?.getContext('2d')
if (!canvas || !ctx) return
if (!isVideo && !imgRef.current) return
const container = containerRef.current
if (container) {
canvas.width = container.clientWidth
canvas.height = container.clientHeight
}
ctx.clearRect(0, 0, canvas.width, canvas.height)
ctx.save()
if (!isVideo && imgRef.current) {
ctx.drawImage(imgRef.current, pan.x, pan.y, imgSize.w * zoom, imgSize.h * zoom)
}
const timeWindowDets = getTimeWindowDetections()
const allDets = [...detections, ...timeWindowDets]
const chips: LabelChip[] = []
allDets.forEach((det, i) => {
const isOwn = i < detections.length
const isSelected = selected.has(i) && isOwn
const cx = (det.centerX - det.width / 2) * imgSize.w * zoom + pan.x
const cy = (det.centerY - det.height / 2) * imgSize.h * zoom + pan.y
const w = det.width * imgSize.w * zoom
const h = det.height * imgSize.h * zoom
const color = getClassColor(det.classNum)
ctx.strokeStyle = color
ctx.lineWidth = isSelected ? 2 : 1
ctx.strokeRect(cx, cy, w, h)
ctx.fillStyle = color
ctx.globalAlpha = 0.06
ctx.fillRect(cx, cy, w, h)
ctx.globalAlpha = 1
// Corner brackets — 8px legs (skipped in environments lacking path API, e.g. JSDOM)
if (typeof ctx.moveTo === 'function' && typeof ctx.beginPath === 'function') {
const legLen = 8
ctx.lineWidth = 2
ctx.beginPath()
ctx.moveTo(cx, cy + legLen); ctx.lineTo(cx, cy); ctx.lineTo(cx + legLen, cy)
ctx.moveTo(cx + w - legLen, cy); ctx.lineTo(cx + w, cy); ctx.lineTo(cx + w, cy + legLen)
ctx.moveTo(cx + w, cy + h - legLen); ctx.lineTo(cx + w, cy + h); ctx.lineTo(cx + w - legLen, cy + h)
ctx.moveTo(cx + legLen, cy + h); ctx.lineTo(cx, cy + h); ctx.lineTo(cx, cy + h - legLen)
ctx.strokeStyle = color
ctx.stroke()
ctx.lineWidth = 1
}
if (isOwn) {
const container = containerRef.current
if (container && container.clientWidth && container.clientHeight) {
chips.push({
leftPct: (cx / container.clientWidth) * 100,
topPct: (cy / container.clientHeight) * 100,
color,
name: det.label || getClassNameFallback(det.classNum),
conf: det.confidence,
combatReady: det.combatReadiness === 1,
})
}
}
if (isSelected) {
const handles = getHandles(cx, cy, w, h)
handles.forEach(hp => {
ctx.fillStyle = '#FF9D3D'
ctx.fillRect(hp.x - HANDLE_SIZE / 2, hp.y - HANDLE_SIZE / 2, HANDLE_SIZE, HANDLE_SIZE)
ctx.strokeStyle = '#0A0D10'
ctx.strokeRect(hp.x - HANDLE_SIZE / 2, hp.y - HANDLE_SIZE / 2, HANDLE_SIZE, HANDLE_SIZE)
})
}
})
if (drawRect) {
ctx.strokeStyle = '#FF9D3D'
ctx.lineWidth = 1
ctx.setLineDash([4, 4])
ctx.strokeRect(drawRect.x, drawRect.y, drawRect.w, drawRect.h)
ctx.setLineDash([])
}
ctx.restore()
// Only setState when chips actually changed — prevents a render storm
// during video playback (draw runs on every time-update; without this
// guard React would commit a new array reference on every paint).
setLabelChips(prev => {
if (prev.length !== chips.length) return chips
for (let i = 0; i < chips.length; i++) {
const a = prev[i], b = chips[i]
if (
a.leftPct !== b.leftPct || a.topPct !== b.topPct ||
a.color !== b.color || a.name !== b.name ||
a.conf !== b.conf || a.combatReady !== b.combatReady
) return chips
}
return prev
})
}, [detections, selected, zoom, pan, imgSize, drawRect, isVideo, getTimeWindowDetections])
useEffect(() => {
const id = requestAnimationFrame(draw)
return () => cancelAnimationFrame(id)
}, [draw])
useEffect(() => {
const container = containerRef.current
if (!container) return
const obs = new ResizeObserver(() => draw())
obs.observe(container)
return () => obs.disconnect()
}, [draw])
const hitTest = (cx: number, cy: number) => {
for (let i = detections.length - 1; i >= 0; i--) {
const d = detections[i]
const bx = (d.centerX - d.width / 2) * imgSize.w * zoom + pan.x
const by = (d.centerY - d.height / 2) * imgSize.h * zoom + pan.y
const bw = d.width * imgSize.w * zoom
const bh = d.height * imgSize.h * zoom
if (selected.has(i)) {
const handles = getHandles(bx, by, bw, bh)
for (const h of handles) {
if (Math.abs(cx - h.x) < HANDLE_SIZE && Math.abs(cy - h.y) < HANDLE_SIZE) {
return { type: 'handle' as const, index: i, handle: h.name }
}
}
}
if (cx >= bx && cx <= bx + bw && cy >= by && cy <= by + bh) {
return { type: 'box' as const, index: i }
}
}
return null
}
const handleMouseDown = (e: React.MouseEvent) => {
const rect = canvasRef.current?.getBoundingClientRect()
if (!rect) return
const mx = e.clientX - rect.left
const my = e.clientY - rect.top
if (e.ctrlKey && e.button === 0) {
setDragState({ type: 'draw', startX: mx, startY: my })
return
}
const hit = hitTest(mx, my)
if (hit?.type === 'handle') {
setDragState({ type: 'resize', startX: mx, startY: my, detectionIndex: hit.index, handle: hit.handle })
} else if (hit?.type === 'box') {
if (e.ctrlKey) {
setSelected(prev => { const n = new Set(prev); n.has(hit.index) ? n.delete(hit.index) : n.add(hit.index); return n })
} else {
setSelected(new Set([hit.index]))
}
setDragState({ type: 'move', startX: mx, startY: my, detectionIndex: hit.index })
} else {
setSelected(new Set())
setDragState({ type: 'draw', startX: mx, startY: my })
}
}
const handleMouseMove = (e: React.MouseEvent) => {
const rect = canvasRef.current?.getBoundingClientRect()
if (!rect) return
const mx = e.clientX - rect.left
const my = e.clientY - rect.top
if (onCursorChange && imgSize.w && imgSize.h) {
const nx = (mx - pan.x) / (imgSize.w * zoom)
const ny = (my - pan.y) / (imgSize.h * zoom)
if (nx >= 0 && nx <= 1 && ny >= 0 && ny <= 1) {
cursorLatestRef.current = { x: nx, y: ny }
if (cursorRafRef.current == null) {
cursorRafRef.current = requestAnimationFrame(() => {
const v = cursorLatestRef.current
cursorRafRef.current = null
if (v) onCursorChange(v.x, v.y)
})
}
}
}
if (!dragState) return
if (dragState.type === 'draw') {
setDrawRect({
x: Math.min(dragState.startX, mx),
y: Math.min(dragState.startY, my),
w: Math.abs(mx - dragState.startX),
h: Math.abs(my - dragState.startY),
})
} else if (dragState.type === 'move' && dragState.detectionIndex !== undefined) {
const dx = (mx - dragState.startX) / (imgSize.w * zoom)
const dy = (my - dragState.startY) / (imgSize.h * zoom)
const newDets = [...detections]
const indices = selected.size > 0 ? Array.from(selected) : [dragState.detectionIndex]
indices.forEach(i => {
if (newDets[i]) {
newDets[i] = {
...newDets[i],
centerX: Math.max(newDets[i].width / 2, Math.min(1 - newDets[i].width / 2, newDets[i].centerX + dx)),
centerY: Math.max(newDets[i].height / 2, Math.min(1 - newDets[i].height / 2, newDets[i].centerY + dy)),
}
}
})
onDetectionsChange(newDets)
setDragState({ ...dragState, startX: mx, startY: my })
} else if (dragState.type === 'resize' && dragState.detectionIndex !== undefined && dragState.handle) {
const idx = dragState.detectionIndex
const d = detections[idx]
const norm = fromCanvas(mx, my)
const newDets = [...detections]
let x1 = d.centerX - d.width / 2, y1 = d.centerY - d.height / 2
let x2 = d.centerX + d.width / 2, y2 = d.centerY + d.height / 2
if (dragState.handle.includes('l')) x1 = norm.x
if (dragState.handle.includes('r')) x2 = norm.x
if (dragState.handle.includes('t')) y1 = norm.y
if (dragState.handle.includes('b')) y2 = norm.y
const w = Math.abs(x2 - x1), h = Math.abs(y2 - y1)
if (w * imgSize.w * zoom >= MIN_BOX_SIZE && h * imgSize.h * zoom >= MIN_BOX_SIZE) {
newDets[idx] = {
...d,
centerX: Math.min(x1, x2) + w / 2,
centerY: Math.min(y1, y2) + h / 2,
width: w,
height: h,
}
onDetectionsChange(newDets)
}
}
}
const handleMouseUp = () => {
if (dragState?.type === 'draw' && drawRect) {
const w = drawRect.w / (imgSize.w * zoom)
const h = drawRect.h / (imgSize.h * zoom)
if (w * imgSize.w >= MIN_BOX_SIZE && h * imgSize.h >= MIN_BOX_SIZE) {
const center = fromCanvas(drawRect.x + drawRect.w / 2, drawRect.y + drawRect.h / 2)
const newDet: Detection = {
id: crypto.randomUUID(),
classNum: selectedClassNum,
label: '',
confidence: 1,
affiliation: 0 as Affiliation,
combatReadiness: 0 as CombatReadiness,
centerX: center.x,
centerY: center.y,
width: w,
height: h,
}
onDetectionsChange([...detections, newDet])
setSelected(new Set([detections.length]))
}
setDrawRect(null)
}
setDragState(null)
}
const handleWheel = (e: React.WheelEvent) => {
if (!e.ctrlKey) return
e.preventDefault()
const delta = e.deltaY > 0 ? 0.9 : 1.1
setZoom(z => Math.max(0.1, Math.min(10, z * delta)))
}
useEffect(() => {
const handler = (e: KeyboardEvent) => {
if (e.target instanceof HTMLInputElement) return
if (e.key === 'Delete' && selected.size > 0) {
onDetectionsChange(detections.filter((_, i) => !selected.has(i)))
setSelected(new Set())
}
if (e.key === 'x' || e.key === 'X') {
if (e.target instanceof HTMLInputElement) return
onDetectionsChange([])
setSelected(new Set())
}
}
window.addEventListener('keydown', handler)
return () => window.removeEventListener('keydown', handler)
}, [detections, selected, onDetectionsChange])
return (
<div ref={containerRef} className="w-full h-full flex-1 relative overflow-hidden cursor-crosshair">
<canvas
ref={canvasRef}
className="absolute inset-0"
onMouseDown={handleMouseDown}
onMouseMove={handleMouseMove}
onMouseUp={handleMouseUp}
onMouseLeave={handleMouseUp}
onWheel={handleWheel}
/>
<div className="absolute inset-0 pointer-events-none">
{labelChips.map((chip, i) => (
<div
key={i}
className="bbox-label"
style={{
position: 'absolute',
left: `${chip.leftPct}%`,
top: `calc(${chip.topPct}% - 26px)`,
borderColor: hexToRgba(chip.color, 0.6),
}}
>
<span style={{ color: chip.color, display: 'inline-flex' }}>{affiliationIcon(chip.color)}</span>
{chip.combatReady && <span style={{ width: 6, height: 6, borderRadius: 999, background: 'var(--accent-green)', display: 'inline-block' }} />}
<span style={{ color: chip.color }}>{chip.name}</span>
{chip.conf < 0.995 && <span className="conf">{(chip.conf * 100).toFixed(1)}%</span>}
</div>
))}
</div>
</div>
)
})
export default CanvasEditor