hitbox-tool: per-frame collision boxes from sprite sheet alpha -> JSON

- parses the spritec _WxH_CxR naming convention incl. auto-detection of
  integer-upscaled sheets (256x64 named 8x8_4x1 -> 64x64 cells)
- tight alpha bbox per frame with threshold/shrink/pad tuning, row-major
  indices, empty-frame flags; boxes relative to frame origin
- show command draws frames + box outline in the terminal for verification
- go tests: scanning, threshold, shrink/pad clamping, name parsing,
  scale inference, JSON roundtrip

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmdG9GqfSWCzts7AkDwRDh
This commit is contained in:
2026-07-14 07:12:05 +02:00
parent dc78c791df
commit f8924d89f7
7 changed files with 698 additions and 0 deletions

66
hitbox-tool/README.md Normal file
View File

@@ -0,0 +1,66 @@
# hitbox-tool (`hitbox`)
Scans **sprite sheet PNGs** and writes per-frame **collision boxes as
JSON**, computed from the alpha channel. Closes the loop with
[`pixel-sprite-maker`](../pixel-sprite-maker/): render a sheet with
`spritec`, scan it with `hitbox`, and your game gets both graphics and
hitboxes without a human drawing rectangles. Go, zero dependencies,
single static binary.
## Build
```bash
# Arch/Garuda: sudo pacman -S go
cd hitbox-tool
go build -o build/hitbox . # or the VS Code task "build hitbox-tool"
go test ./...
```
## Usage
```bash
# spritec-named sheets need no flags — layout is parsed from the name,
# and integer upscales are auto-detected (a _8x8_4x1 sheet that is
# 256x64 px was rendered at --scale 8, so cells are 64x64):
hitbox scan walk_8x8_4x1.png -o walk.hitbox.json
hitbox scan boss.png --cell 32x32 -o boss.json # explicit frame size
hitbox scan portrait.png # whole image = one frame, JSON to stdout
hitbox show walk_8x8_4x1.png --frame 2 # verify visually in the terminal
# tuning:
--threshold 128 # ignore faint pixels (alpha < 128)
--shrink 1 # tighter, more forgiving hitboxes (n px per side)
--pad 2 # generous hitboxes (e.g. pickups)
```
## Output
```json
{
"image": "walk_8x8_4x1.png",
"cellW": 8, "cellH": 8, "cols": 4, "rows": 1,
"alphaThreshold": 1,
"frames": [
{ "index": 0, "col": 0, "row": 0, "empty": false,
"box": { "x": 1, "y": 0, "w": 5, "h": 8 } },
{ "index": 1, "col": 1, "row": 0, "empty": true, "box": null }
]
}
```
- Boxes are **relative to each frame's top-left corner**.
- Frame order is row-major (`index = row * cols + col`), matching
spritec sheets.
- Cells with no solid pixels get `"empty": true`.
In game code:
```
hit = px >= frameX + box.x && px < frameX + box.x + box.w
&& py >= frameY + box.y && py < frameY + box.y + box.h
```
`hitbox show` draws each frame with `#` for solid pixels and `+` for the
box outline, so an agent can verify the result without an image viewer.

View File

@@ -0,0 +1,58 @@
{
"image": "examples/walk_8x8_4x1.png",
"cellW": 64,
"cellH": 64,
"cols": 4,
"rows": 1,
"alphaThreshold": 1,
"frames": [
{
"index": 0,
"col": 0,
"row": 0,
"empty": false,
"box": {
"x": 8,
"y": 0,
"w": 40,
"h": 64
}
},
{
"index": 1,
"col": 1,
"row": 0,
"empty": false,
"box": {
"x": 8,
"y": 0,
"w": 40,
"h": 64
}
},
{
"index": 2,
"col": 2,
"row": 0,
"empty": false,
"box": {
"x": 8,
"y": 0,
"w": 40,
"h": 64
}
},
{
"index": 3,
"col": 3,
"row": 0,
"empty": false,
"box": {
"x": 8,
"y": 0,
"w": 40,
"h": 64
}
}
]
}

Binary file not shown.

After

Width:  |  Height:  |  Size: 508 B

3
hitbox-tool/go.mod Normal file
View File

@@ -0,0 +1,3 @@
module gitea.brasse-pc.eu/brasse/agent-tools/hitbox-tool
go 1.24

View File

@@ -0,0 +1,230 @@
// Package hitbox computes per-frame collision boxes from sprite sheet
// images by scanning the alpha channel.
package hitbox
import (
"encoding/json"
"fmt"
"image"
"image/png"
"io"
"os"
"path/filepath"
"regexp"
"strconv"
)
// Box is a rectangle relative to its frame's top-left corner.
type Box struct {
X int `json:"x"`
Y int `json:"y"`
W int `json:"w"`
H int `json:"h"`
}
// Frame is the scan result for one sheet cell.
type Frame struct {
Index int `json:"index"`
Col int `json:"col"`
Row int `json:"row"`
Empty bool `json:"empty"`
Box *Box `json:"box"` // nil when Empty
}
// Sheet is the full scan result; the JSON deliverable.
type Sheet struct {
Image string `json:"image"`
CellW int `json:"cellW"`
CellH int `json:"cellH"`
Cols int `json:"cols"`
Rows int `json:"rows"`
Threshold int `json:"alphaThreshold"`
Frames []Frame `json:"frames"`
}
// layoutRe matches the spritec sheet naming convention:
// <base>_<cellW>x<cellH>_<cols>x<rows>.<ext>
var layoutRe = regexp.MustCompile(`_(\d+)x(\d+)_(\d+)x(\d+)\.[A-Za-z]+$`)
// LayoutFromName extracts cell size and grid from a spritec-style file
// name. ok is false when the name doesn't follow the convention.
func LayoutFromName(path string) (cellW, cellH, cols, rows int, ok bool) {
m := layoutRe.FindStringSubmatch(filepath.Base(path))
if m == nil {
return 0, 0, 0, 0, false
}
cellW, _ = strconv.Atoi(m[1])
cellH, _ = strconv.Atoi(m[2])
cols, _ = strconv.Atoi(m[3])
rows, _ = strconv.Atoi(m[4])
return cellW, cellH, cols, rows, true
}
// InferScale detects integer-upscaled sheets: when the image is exactly
// s times bigger than the name-declared layout (both axes, s >= 1), the
// real cell size is cell*s. Returns 0 when the layout doesn't fit.
func InferScale(imgW, imgH, cellW, cellH, cols, rows int) int {
baseW, baseH := cellW*cols, cellH*rows
if baseW <= 0 || baseH <= 0 || imgW%baseW != 0 || imgH%baseH != 0 {
return 0
}
s := imgW / baseW
if s < 1 || imgH/baseH != s {
return 0
}
return s
}
// LoadPNG reads a PNG image from disk.
func LoadPNG(path string) (image.Image, error) {
f, err := os.Open(path)
if err != nil {
return nil, err
}
defer f.Close()
img, err := png.Decode(f)
if err != nil {
return nil, fmt.Errorf("%s: %w (only PNG is supported — sheets need an alpha channel)", path, err)
}
return img, nil
}
// Scan computes a tight box per frame: the smallest rectangle covering
// every pixel with alpha >= threshold. shrink/pad (in pixels) contract or
// expand each box afterwards, clamped to the cell.
func Scan(img image.Image, name string, cellW, cellH, threshold, shrink, pad int) (*Sheet, error) {
b := img.Bounds()
if cellW <= 0 || cellH <= 0 {
cellW, cellH = b.Dx(), b.Dy() // whole image = one frame
}
if b.Dx()%cellW != 0 || b.Dy()%cellH != 0 {
return nil, fmt.Errorf("image is %dx%d which is not divisible by cell %dx%d",
b.Dx(), b.Dy(), cellW, cellH)
}
if threshold < 1 || threshold > 255 {
return nil, fmt.Errorf("alpha threshold %d out of range 1-255", threshold)
}
cols, rows := b.Dx()/cellW, b.Dy()/cellH
sheet := &Sheet{
Image: name, CellW: cellW, CellH: cellH,
Cols: cols, Rows: rows, Threshold: threshold,
}
for row := 0; row < rows; row++ {
for col := 0; col < cols; col++ {
fr := Frame{Index: row*cols + col, Col: col, Row: row}
minX, minY := cellW, cellH
maxX, maxY := -1, -1
for y := 0; y < cellH; y++ {
for x := 0; x < cellW; x++ {
_, _, _, a := img.At(b.Min.X+col*cellW+x, b.Min.Y+row*cellH+y).RGBA()
if int(a>>8) >= threshold {
if x < minX {
minX = x
}
if y < minY {
minY = y
}
if x > maxX {
maxX = x
}
if y > maxY {
maxY = y
}
}
}
}
if maxX < 0 {
fr.Empty = true
} else {
box := Box{X: minX, Y: minY, W: maxX - minX + 1, H: maxY - minY + 1}
box = adjust(box, shrink-pad, cellW, cellH)
fr.Box = &box
}
sheet.Frames = append(sheet.Frames, fr)
}
}
return sheet, nil
}
// adjust contracts the box by delta px on every side (negative delta
// expands), clamped to the cell and to a minimum size of 1x1.
func adjust(b Box, delta, cellW, cellH int) Box {
b.X += delta
b.Y += delta
b.W -= 2 * delta
b.H -= 2 * delta
if b.X < 0 {
b.W += b.X
b.X = 0
}
if b.Y < 0 {
b.H += b.Y
b.Y = 0
}
if b.X+b.W > cellW {
b.W = cellW - b.X
}
if b.Y+b.H > cellH {
b.H = cellH - b.Y
}
if b.W < 1 {
b.W = 1
if b.X > cellW-1 {
b.X = cellW - 1
}
}
if b.H < 1 {
b.H = 1
if b.Y > cellH-1 {
b.Y = cellH - 1
}
}
return b
}
// WriteJSON encodes the sheet as indented JSON.
func (s *Sheet) WriteJSON(w io.Writer) error {
enc := json.NewEncoder(w)
enc.SetIndent("", " ")
return enc.Encode(s)
}
// Ascii draws one frame with '#' for solid pixels and '+' for the box
// outline, so results can be verified in a terminal.
func Ascii(img image.Image, s *Sheet, index, threshold int) string {
if index < 0 || index >= len(s.Frames) {
return "(no such frame)\n"
}
fr := s.Frames[index]
b := img.Bounds()
out := make([]rune, 0, (s.CellW+1)*s.CellH)
onBoxEdge := func(x, y int) bool {
if fr.Box == nil {
return false
}
bx := fr.Box
inX := x >= bx.X && x < bx.X+bx.W
inY := y >= bx.Y && y < bx.Y+bx.H
edgeX := x == bx.X || x == bx.X+bx.W-1
edgeY := y == bx.Y || y == bx.Y+bx.H-1
return (inX && inY) && (edgeX || edgeY)
}
for y := 0; y < s.CellH; y++ {
for x := 0; x < s.CellW; x++ {
_, _, _, a := img.At(b.Min.X+fr.Col*s.CellW+x, b.Min.Y+fr.Row*s.CellH+y).RGBA()
solid := int(a>>8) >= threshold
switch {
case solid && onBoxEdge(x, y):
out = append(out, '#')
case solid:
out = append(out, '#')
case onBoxEdge(x, y):
out = append(out, '+')
default:
out = append(out, '.')
}
}
out = append(out, '\n')
}
return string(out)
}

View File

@@ -0,0 +1,137 @@
package hitbox
import (
"bytes"
"encoding/json"
"image"
"image/color"
"testing"
)
// sheet4 builds a 2x1 sheet of 4x4 cells: frame 0 has a 2x2 blob at
// (1,1); frame 1 is empty.
func sheet4() image.Image {
img := image.NewNRGBA(image.Rect(0, 0, 8, 4))
for y := 1; y <= 2; y++ {
for x := 1; x <= 2; x++ {
img.SetNRGBA(x, y, color.NRGBA{255, 0, 0, 255})
}
}
return img
}
func TestScanTightBox(t *testing.T) {
s, err := Scan(sheet4(), "test.png", 4, 4, 1, 0, 0)
if err != nil {
t.Fatal(err)
}
if s.Cols != 2 || s.Rows != 1 || len(s.Frames) != 2 {
t.Fatalf("layout %dx%d frames=%d", s.Cols, s.Rows, len(s.Frames))
}
f0 := s.Frames[0]
if f0.Empty || f0.Box == nil {
t.Fatal("frame 0 should have a box")
}
if *f0.Box != (Box{X: 1, Y: 1, W: 2, H: 2}) {
t.Errorf("frame 0 box = %+v", *f0.Box)
}
f1 := s.Frames[1]
if !f1.Empty || f1.Box != nil {
t.Errorf("frame 1 should be empty, got %+v", f1)
}
}
func TestScanWholeImageDefault(t *testing.T) {
s, err := Scan(sheet4(), "x.png", 0, 0, 1, 0, 0)
if err != nil {
t.Fatal(err)
}
if len(s.Frames) != 1 || s.CellW != 8 || s.CellH != 4 {
t.Errorf("whole-image scan: %+v", s)
}
}
func TestScanErrors(t *testing.T) {
if _, err := Scan(sheet4(), "x.png", 3, 4, 1, 0, 0); err == nil {
t.Error("non-divisible cell size should error")
}
if _, err := Scan(sheet4(), "x.png", 4, 4, 0, 0, 0); err == nil {
t.Error("threshold 0 should error")
}
}
func TestShrinkAndPad(t *testing.T) {
img := image.NewNRGBA(image.Rect(0, 0, 4, 4))
for y := 0; y < 4; y++ {
for x := 0; x < 4; x++ {
img.SetNRGBA(x, y, color.NRGBA{0, 0, 0, 255})
}
}
s, _ := Scan(img, "x.png", 4, 4, 1, 1, 0) // shrink 1
if *s.Frames[0].Box != (Box{X: 1, Y: 1, W: 2, H: 2}) {
t.Errorf("shrunk box = %+v", *s.Frames[0].Box)
}
s, _ = Scan(img, "x.png", 4, 4, 1, 0, 3) // pad clamps to cell
if *s.Frames[0].Box != (Box{X: 0, Y: 0, W: 4, H: 4}) {
t.Errorf("padded box = %+v", *s.Frames[0].Box)
}
s, _ = Scan(img, "x.png", 4, 4, 1, 10, 0) // over-shrink -> min 1x1
if s.Frames[0].Box.W < 1 || s.Frames[0].Box.H < 1 {
t.Errorf("over-shrunk box = %+v", *s.Frames[0].Box)
}
}
func TestThreshold(t *testing.T) {
img := image.NewNRGBA(image.Rect(0, 0, 2, 1))
img.SetNRGBA(0, 0, color.NRGBA{0, 0, 0, 100})
img.SetNRGBA(1, 0, color.NRGBA{0, 0, 0, 200})
s, _ := Scan(img, "x.png", 2, 1, 150, 0, 0)
if *s.Frames[0].Box != (Box{X: 1, Y: 0, W: 1, H: 1}) {
t.Errorf("threshold box = %+v", *s.Frames[0].Box)
}
s, _ = Scan(img, "x.png", 2, 1, 50, 0, 0)
if *s.Frames[0].Box != (Box{X: 0, Y: 0, W: 2, H: 1}) {
t.Errorf("low-threshold box = %+v", *s.Frames[0].Box)
}
}
func TestLayoutFromName(t *testing.T) {
w, h, c, r, ok := LayoutFromName("/tmp/walk_8x8_4x1.png")
if !ok || w != 8 || h != 8 || c != 4 || r != 1 {
t.Errorf("parsed %d %d %d %d ok=%v", w, h, c, r, ok)
}
if _, _, _, _, ok := LayoutFromName("plain.png"); ok {
t.Error("plain.png should not match")
}
}
func TestInferScale(t *testing.T) {
// walk_8x8_4x1.png rendered at scale 8 -> 256x64
if s := InferScale(256, 64, 8, 8, 4, 1); s != 8 {
t.Errorf("scale = %d, want 8", s)
}
if s := InferScale(32, 8, 8, 8, 4, 1); s != 1 {
t.Errorf("unscaled = %d, want 1", s)
}
if s := InferScale(250, 64, 8, 8, 4, 1); s != 0 {
t.Errorf("non-divisible = %d, want 0", s)
}
if s := InferScale(256, 32, 8, 8, 4, 1); s != 0 {
t.Errorf("axis mismatch = %d, want 0", s)
}
}
func TestJSONShape(t *testing.T) {
s, _ := Scan(sheet4(), "test.png", 4, 4, 1, 0, 0)
var buf bytes.Buffer
if err := s.WriteJSON(&buf); err != nil {
t.Fatal(err)
}
var back Sheet
if err := json.Unmarshal(buf.Bytes(), &back); err != nil {
t.Fatalf("invalid json: %v", err)
}
if back.Frames[0].Box.W != 2 {
t.Errorf("json roundtrip box = %+v", back.Frames[0].Box)
}
}

204
hitbox-tool/main.go Normal file
View File

@@ -0,0 +1,204 @@
// hitbox scans sprite sheet PNGs and writes per-frame collision boxes
// as JSON, using the alpha channel to find each frame's solid pixels.
package main
import (
"flag"
"fmt"
"os"
"strconv"
"strings"
"gitea.brasse-pc.eu/brasse/agent-tools/hitbox-tool/hitbox"
)
var version = "dev"
const usage = `hitbox - collision box annotator for sprite sheets
Usage:
hitbox scan <sheet.png> [flags] compute per-frame boxes -> JSON
hitbox show <sheet.png> [flags] draw frames + boxes in the terminal
hitbox version
Scan flags:
--cell <WxH> frame size, e.g. 8x8. Default: parsed from the
spritec naming convention <name>_<W>x<H>_<C>x<R>.png;
if neither is given the whole image is one frame.
--threshold <n> alpha 1-255 that counts as solid (default 1)
--shrink <n> contract every box by n px per side (forgiving hits)
--pad <n> expand every box by n px per side
-o <path> write JSON here (default: stdout)
Show flags: --cell, --threshold, plus
--frame <n> only this frame index (default: all)
Boxes are relative to each frame's top-left corner. Frame indices are
row-major (index = row * cols + col), matching spritec sheets.
In game code: hit if (px,py) inside (frameX + box.x, frameY + box.y,
box.w, box.h).
`
func main() {
if len(os.Args) < 2 {
fmt.Print(usage)
os.Exit(2)
}
switch os.Args[1] {
case "scan":
cmdScan(os.Args[2:])
case "show":
cmdShow(os.Args[2:])
case "version", "--version", "-v":
fmt.Println("hitbox", version)
case "help", "--help", "-h":
fmt.Print(usage)
default:
die("unknown command %q — run 'hitbox help'", os.Args[1])
}
}
// parseInterspersed lets flags appear before or after positional args.
func parseInterspersed(fs *flag.FlagSet, args []string) {
var flags, pos []string
for i := 0; i < len(args); i++ {
a := args[i]
if len(a) > 1 && a[0] == '-' {
flags = append(flags, a)
name := strings.TrimLeft(a, "-")
if !strings.Contains(name, "=") {
f := fs.Lookup(name)
isBool := false
if f != nil {
if bv, ok := f.Value.(interface{ IsBoolFlag() bool }); ok && bv.IsBoolFlag() {
isBool = true
}
}
if !isBool && i+1 < len(args) {
i++
flags = append(flags, args[i])
}
}
} else {
pos = append(pos, a)
}
}
fs.Parse(append(flags, pos...))
}
// parseCell resolves the frame size from --cell, or from the spritec
// naming convention (auto-detecting integer upscales: a sheet named
// _8x8_4x1 that is 256x64 px was rendered at scale 8, so cells are 64x64).
func parseCell(spec, path string, imgW, imgH int) (int, int, error) {
if spec != "" {
parts := strings.SplitN(strings.ToLower(spec), "x", 2)
if len(parts) == 2 {
w, err1 := strconv.Atoi(parts[0])
h, err2 := strconv.Atoi(parts[1])
if err1 == nil && err2 == nil && w > 0 && h > 0 {
return w, h, nil
}
}
return 0, 0, fmt.Errorf("--cell %q must look like 8x8", spec)
}
if w, h, cols, rows, ok := hitbox.LayoutFromName(path); ok {
if s := hitbox.InferScale(imgW, imgH, w, h, cols, rows); s > 1 {
fmt.Fprintf(os.Stderr, "hitbox: image is %dx the named layout — using %dx%d cells\n", s, w*s, h*s)
return w * s, h * s, nil
}
return w, h, nil
}
return 0, 0, nil // whole image = one frame
}
func cmdScan(args []string) {
fs := flag.NewFlagSet("scan", flag.ExitOnError)
cell := fs.String("cell", "", "frame size WxH")
threshold := fs.Int("threshold", 1, "solid alpha 1-255")
shrink := fs.Int("shrink", 0, "contract boxes n px per side")
pad := fs.Int("pad", 0, "expand boxes n px per side")
out := fs.String("o", "", "output JSON path (default stdout)")
parseInterspersed(fs, args)
if fs.NArg() != 1 {
die("scan takes exactly one PNG file")
}
path := fs.Arg(0)
img, err := hitbox.LoadPNG(path)
if err != nil {
die("%v", err)
}
b := img.Bounds()
cw, ch, err := parseCell(*cell, path, b.Dx(), b.Dy())
if err != nil {
die("%v", err)
}
sheet, err := hitbox.Scan(img, path, cw, ch, *threshold, *shrink, *pad)
if err != nil {
die("%v", err)
}
if *out == "" {
if err := sheet.WriteJSON(os.Stdout); err != nil {
die("%v", err)
}
return
}
f, err := os.Create(*out)
if err != nil {
die("%v", err)
}
defer f.Close()
if err := sheet.WriteJSON(f); err != nil {
die("%v", err)
}
solid := 0
for _, fr := range sheet.Frames {
if !fr.Empty {
solid++
}
}
fmt.Printf("%s (%d frames of %dx%d, %d with pixels)\n",
*out, len(sheet.Frames), sheet.CellW, sheet.CellH, solid)
}
func cmdShow(args []string) {
fs := flag.NewFlagSet("show", flag.ExitOnError)
cell := fs.String("cell", "", "frame size WxH")
threshold := fs.Int("threshold", 1, "solid alpha 1-255")
frame := fs.Int("frame", -1, "frame index (default: all)")
parseInterspersed(fs, args)
if fs.NArg() != 1 {
die("show takes exactly one PNG file")
}
path := fs.Arg(0)
img, err := hitbox.LoadPNG(path)
if err != nil {
die("%v", err)
}
b := img.Bounds()
cw, ch, err := parseCell(*cell, path, b.Dx(), b.Dy())
if err != nil {
die("%v", err)
}
sheet, err := hitbox.Scan(img, path, cw, ch, *threshold, 0, 0)
if err != nil {
die("%v", err)
}
for _, fr := range sheet.Frames {
if *frame >= 0 && fr.Index != *frame {
continue
}
if fr.Empty {
fmt.Printf("frame %d (col %d, row %d): empty\n\n", fr.Index, fr.Col, fr.Row)
continue
}
fmt.Printf("frame %d (col %d, row %d): box x=%d y=%d w=%d h=%d\n",
fr.Index, fr.Col, fr.Row, fr.Box.X, fr.Box.Y, fr.Box.W, fr.Box.H)
fmt.Print(hitbox.Ascii(img, sheet, fr.Index, *threshold))
fmt.Println()
}
}
func die(format string, a ...any) {
fmt.Fprintf(os.Stderr, "hitbox: "+format+"\n", a...)
os.Exit(1)
}