Skip to content

refactor(moh_native): remove unused code #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
dmitriy-myz merged 1 commit into from
Jun 6, 2026
Merged

refactor(moh_native): remove unused code #2

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
93 changes: 5 additions & 88 deletions validitysensor/moh_native.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,7 +11,7 @@
→ BRIEF bit-pack (per-kp 128 binary tests) [brief_pack] BYTE-EXACT
→ orientation (Gaussian-weighted grad histogram) [orient_d920] BYTE-EXACT (60/60)
→ oriented BRIEF descriptor [descriptor] BYTE-EXACT (2026-05-30)
→ [x][y][128-bit desc] × 250 → v30 [build_v30] format known
→ [x][y][128-bit desc] × 250 → v30 [serialize_v30_section] format known

Validation: each stage is checked against the live captures in
$FRIDA_DUMP_DIR (see dev/diff_v30.py). The tiling stage matches `gradin`
Expand Down Expand Up @@ -106,8 +106,6 @@ def tile_image(img):
COS_Q16 = np.round(np.cos(np.deg2rad(np.arange(360))) * 65536).astype(np.int64)
SIN_Q16 = np.round(np.sin(np.deg2rad(np.arange(360))) * 65536).astype(np.int64)

ATAN2_FULLSCALE = np.pi * 65536 # 205887.4 — sub_180003150 Q16-radian full scale


def orient_to_index(orient_q16):
"""Convert kp[+0xc] orient_q16 (Q16 radians, [0, π·65536)) to cos/sin table
Expand Down Expand Up @@ -348,19 +346,6 @@ def subpix_refine_kp(resp, x_int, y_int, scale_shift=0):
return x_q16, y_q16


def subpix_refine_kps(resp, kps, scale_shift=0):
"""sub_18000D5D0 driver — refine all NMS keypoints, CULL failures.
`kps`: list of (score, x_int, y_int) from `nms()`.
Returns: list of (score, x_q16, y_q16). Length ≤ input (failed kps
are removed, exactly like the DLL's sub_18000D570 memmove-down)."""
out = []
for s, x, y in kps:
r = subpix_refine_kp(resp, x, y, scale_shift)
if r is not None:
out.append((s, r[0], r[1]))
return out


# NEXT after NMS: orientation (sub_18000D920) + oriented BRIEF (sub_18000E090).


Expand Down Expand Up @@ -546,13 +531,6 @@ def P(im, kx, ky):
# Header = [u16 tag=4][u16 len=4533][8 zeros].


def tile_origin_yx(i, j, h, w):
"""Top-left (row, col) of tile (i, j) in the 3×3 grid. Re-export of
tile_image's internal `tile_origin` formula so callers can use it
independently of the iteration."""
return tile_origin(i, j, h, w)


def merge_tile_kps_to_global(per_tile_kps, h, w, margin=3):
"""Merge per-tile keypoint lists into a single global list, applying
the DLL's bound check (margin ≤ global_xy < dim-margin).
Expand Down Expand Up @@ -587,38 +565,8 @@ def merge_tile_kps_to_global(per_tile_kps, h, w, margin=3):
return out


def build_v30(global_kps, max_records=250, tag=4, body_len=4533):
"""Build the 18-byte-per-record v30 buffer that goes into a frame
section. Layout matches the captured v30: 12 B header + body_len
payload (~17 B lead-in zeros + N × 18 B records + trailer zeros).

`global_kps`: list of (gx_int, gy_int, ..., desc_16B) from
merge_tile_kps_to_global. Only x, y, and descriptor are used —
orient/quality stay in the per-kp records elsewhere.

Returns: bytes object of (12 + body_len) bytes."""
header = bytes([tag & 0xFF, (tag >> 8) & 0xFF,
body_len & 0xFF, (body_len >> 8) & 0xFF]) + bytes(8)
body = bytearray(body_len)
# 17-byte zero lead-in (matches the captured records — empirical;
# disasm of the v30 packer is pending).
rec_offset = 17
for kp in global_kps[:max_records]:
gx, gy = kp[0], kp[1]
desc = kp[-1] # last field is the 16-byte descriptor
if not isinstance(desc, (bytes, bytearray)):
desc = bytes(desc)
body[rec_offset] = gx & 0xFF
body[rec_offset + 1] = gy & 0xFF
body[rec_offset + 2:rec_offset + 18] = desc[:16]
rec_offset += 18
return header + bytes(body)


# ─── WS-body v30 SECTION serializer ──────────────────────────────────────
# The v30 record area of ONE ws-body section. Distinct from build_v30()
# above, which models the *standalone* per-frame v30 buffer (12-B header +
# 17-B lead-in + body). A ws-body section is a PURE record area: exactly
# The v30 record area of ONE ws-body section: a PURE record area of exactly
# n_slots × 18-byte records, no header/lead-in/trailer.
#
# GROUND-TRUTH CONFIRMED (gdb capture ws_body_1780084103036_23056.bin,
Expand All @@ -630,31 +578,10 @@ def build_v30(global_kps, max_records=250, tag=4, body_len=4533):
V30_SECTION_RECORDS = 250
V30_SECTION_BYTES = V30_SECTION_RECORDS * 18 # 4500
V30_DESC_LEN = 16 # record = [desc:16][x:u8][y:u8]; x,y anchor is +16 into the record
V30_RECORD_LEN = 18 # full v30 record stride
WS_SIZE = 23056 # chip-view WS body size
DEFAULT_SUBTYPE = 0x00f7 # default WinBio finger subtype


def find_v30_regions(ws, min_run=30):
"""Locate each section's v30 record array by detecting long runs of 18-byte
records. Returns the (x,y) ANCHOR offset of each region; the record layout is
[16B desc][x:u8][y:u8], so the record area (first descriptor) starts at
`anchor - V30_DESC_LEN` and the (x,y) bytes are at +16/+17. (Moved from the
retired moh_opencv module.)"""
regions, p, n = [], 0, len(ws)
while p < n - V30_RECORD_LEN * min_run:
good, q = 0, p
while q + 1 < n and 0 < ws[q] <= 112 and ws[q + 1] <= 112:
good += 1
q += V30_RECORD_LEN
if good >= min_run:
regions.append(p)
p += V30_SECTION_RECORDS * V30_RECORD_LEN
else:
p += 1
return regions


def serialize_v30_section(records, n_slots=V30_SECTION_RECORDS):
"""Serialize one ws-body section's v30 record area.

Expand All @@ -666,7 +593,7 @@ def serialize_v30_section(records, n_slots=V30_SECTION_RECORDS):
FIRST), decoded from the v30 emitter sub_1800057e0 (2026-06-01) and
verified: parsing a stored section this way pairs (x,y,desc) 238-248/250
against our single-frame extraction (vs 0/250 for the old [x][y][desc]).
The section's record area starts at (find_v30_regions anchor − 16), so
The section's record area starts at (the v30-region anchor − 16), so
callers must write this buffer at `anchor - V30_DESC_LEN`."""
out = bytearray(n_slots * 18)
for i, rec in enumerate(records):
Expand Down Expand Up @@ -721,16 +648,6 @@ def serialize_v30_section(records, n_slots=V30_SECTION_RECORDS):
# enrollment with GDB_DUMP_DESC_BRIEF=1 to grab them.


def _rotate_sample_pair(gx, gy, cos_q16, sin_q16):
"""E090 inner rotation: takes raw (gx, gy) i32 samples, returns rotated pair.
Bit-exact emulation of the x86 imul/sar sequence at e3cd-e418."""
gx8 = np.int32(gx) >> 8
gy8 = np.int32(gy) >> 8
rgx = (np.int32(cos_q16 * gx8) >> 8) + (np.int32(sin_q16 * gy8) >> 8)
rgy = (np.int32(cos_q16 * gy8) >> 8) - (np.int32(sin_q16 * gx8) >> 8)
return np.int32(rgx), np.int32(rgy)


def desc_sample_rotate(grad_x, grad_y, subpix_x_q16, subpix_y_q16, orient_idx,
N=7):
"""E090 rotation+sampling (stage 1). Returns two int32 arrays of length
Expand Down Expand Up @@ -1009,8 +926,8 @@ def extract_frame_native(image_q16, h=112, w=112,
# captured reference template needs to be supplied. The real descriptors of
# the source template are NOT shipped (zeroed for privacy + clarity).
#
# The v30 record areas live at these fixed offsets in the scaffold (from
# find_v30_regions on fresh.bin); each is 250×18 = 4500 bytes. They're pinned
# The v30 record areas live at these fixed offsets in the scaffold (detected
# on fresh.bin); each is 250×18 = 4500 bytes. They're pinned
# rather than re-detected because the scaffold's record areas are zeroed (so
# the coordinate-run heuristic can't find them).
NATIVE_WS_V30_REGIONS = (309, 4913, 9453, 13993)
Expand Down