steam-frame/DeeJanuz--frametop
1
0 Fork 0
mirror of https://github.com/DeeJanuz/frametop.git synced 2026-10-06 09:00:11 +02:00
Code Issues Packages Projects Releases 4 Wiki Activity

Hand tracking for Frametop's hand cutouts

Browse Source 
fh-camd (camd/) borrows XRService's camera buffers and publishes the tracking
cameras' frames to a shared ring. fh-tracker (trackd/) finds hands in them with
MediaPipe's palm and landmark models on ncnn, triangulates them in 3D, and
publishes them for ft-screens. fh-replay replays recordings offline. tracker/ is
the earlier Python version; tools/ and probes/ hold the checks and experiments.

As of this commit: crop contrast defaults to CLAHE for the palm search and plain
crops for the landmarks, --swap-sides works around fh-camd naming the side
cameras backwards after some XRService restarts (tools/check_sides.py detects
it), and --record-only, --with-dark, --cpus and --keep-presence support the
bright-light and CPU-placement tests.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
DeeJanuzandClaude Opus 5.5 committed 2026-09-29 22:36:48 -06:00
commit 067a03ce38
42 files changed
+6321
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

trackd/replay.cpp
+240
View File
@@ -0,0 +1,240 @@
// fh-replay: run a recording (fh-tracker --record) through the tracker offline, with the
// live scheduling, and report how well it kept the hands.
//
// fh-replay DIR [--oracle N] [--slow F] [--timeline FILE] [--threads N] [--models DIR]
// [--from S] [--to S] [--contrast MODE|PALM/HAND] (clahe[:CLIP], none, stretch)
//
// --oracle N: every N-th set, also search every tile of every camera (slow), to see
// which hands were there to find. Compares that with what the tracker had.
// --slow F: the live tracker skips the sets that arrive while it's busy; replay takes
// each step's time here times F as the busy time (the headset is busier live).
// --cost: instead of timing the steps, charge each round of model calls what it
// typically costs live (10 ms landmarks, 18 ms palms): repeatable results.
// --timeline: per processed set, a line per hand (time, id, side, views, wrist) and per view
// (hand, camera, presence, next crop, set index).
// --keep-presence P: landmark presence a tracked view needs to stay (default 0.5, as new ones).
// --contrast: how the palm search's and the landmark model's crops are equalized
// (default clahe:2/none, as fh-tracker).
#include "record.h"
#include "tracker.h"
#include <algorithm>
#include <chrono>
#include <cstdio>
#include <cstring>
#include <map>
#include <set>
#include <string>
#include <vector>
namespace {
struct Track {
double first = 0, last = 0;
int sets = 0, left = 0;
// the last two palm positions (raw, smoothed) and times, for the jitter measure
V3 raw[2]{}, sm[2]{};
double t[2]{};
int line = 0; // updates on the current unbroken run
};
double median(std::vector<double> v) {
if (v.empty()) return 0;
std::nth_element(v.begin(), v.begin() + v.size() / 2, v.end());
return v[v.size() / 2];
}
} // namespace
int main(int argc, char **argv) {
if (argc < 2 || argv[1][0] == '-') {
std::printf("usage: %s DIR [--oracle N] [--slow F] [--timeline FILE] [--threads N] [--models DIR] [--from S] [--to S]\n", argv[0]);
return 1;
}
const std::string dir = argv[1];
int oracle = 0, threads = 2;
double slow = 1.0, from = 0, to = 1e9;
bool cost = false;
Contrast palm_contrast, hand_contrast{Contrast::None}; // as fh-tracker's
double keep_presence = 0.5; // landmark presence a tracked view needs to stay
std::string timeline, models = std::string(argv[0]).substr(0, std::string(argv[0]).rfind('/') + 1) + "../models/ncnn";
for (int i = 2; i < argc; ++i) {
const std::string a = argv[i];
const bool more = i + 1 < argc;
if (a == "--oracle" && more) oracle = std::atoi(argv[++i]);
else if (a == "--slow" && more) slow = std::atof(argv[++i]);
else if (a == "--timeline" && more) timeline = argv[++i];
else if (a == "--threads" && more) threads = std::atoi(argv[++i]);
else if (a == "--models" && more) models = argv[++i];
else if (a == "--cost") cost = true;
else if (a == "--keep-presence" && more) keep_presence = std::atof(argv[++i]);
else if (a == "--contrast" && more) {
if (!Contrast::parse_pair(argv[++i], palm_contrast, hand_contrast))
return std::fprintf(stderr, "--contrast MODE or PALM/HAND, each clahe[:CLIP]|none|stretch\n"), 1;
}
else if (a == "--from" && more) from = std::atof(argv[++i]);
else if (a == "--to" && more) to = std::atof(argv[++i]);
else return std::fprintf(stderr, "unknown option %s\n", a.c_str()), 1;
}
std::string err;
std::map<std::string, Camera> calib;
Nets nets;
SetReader in;
if (!load_calibration(calib, err) || !nets.load(models, false, err) || !in.open(dir, err))
return std::fprintf(stderr, "%s\n", err.c_str()), 1;
nets.set_contrast(palm_contrast, hand_contrast);
FILE *tl = timeline.empty() ? nullptr : std::fopen(timeline.c_str(), "w");
std::vector<fh_set_cam_t> cams;
std::vector<std::vector<uint8_t>> px;
if (!in.next(cams, px)) return std::fprintf(stderr, "%s: no sets\n", dir.c_str()), 1;
std::map<std::string, Camera> used;
for (auto &c : cams)
if (calib.count(c.name)) used[c.name] = calib[c.name];
Pool pool(threads, {2, 3, 4});
Tracker tracker(used, nets, pool);
tracker.set_keep_presence(keep_presence);
uint64_t t0 = 0, busy_until = 0, next_ns = 0;
int index = -1; // of the set in the recording
int nsets = 0, processed = 0, left = 0, right = 0, both = 0, hist[3] = {};
std::map<int, Track> tracks;
std::vector<const Hand *> last_out;
// oracle: sets where a side's hand was findable, and where the tracker had it then
int o_sets = 0, o_left = 0, o_right = 0, o_left_hit = 0, o_right_hit = 0, o_left_extra = 0, o_right_extra = 0;
std::map<std::string, int> o_by_cam;
double busy_ms = 0;
// jitter: how far each update's palm is from a straight line through the last two,
// mm (steady motion cancels out; what's left is noise and real acceleration)
std::vector<double> jit_raw, jit_sm;
int near_face = 0, hand_updates = 0; // published palms within 20 cm of the eyes
do {
std::map<std::string, Image> images;
uint64_t t = UINT64_MAX;
for (size_t i = 0; i < cams.size(); ++i) {
if (!used.count(cams[i].name)) continue;
images[cams[i].name] = {px[i].data(), int(cams[i].width), int(cams[i].height), int(cams[i].width)};
t = std::min(t, cams[i].capture_ns);
}
if (!t0) t0 = t;
const double ts = (t - t0) / 1e9;
++index;
if (ts < from) continue;
if (ts > to) break;
++nsets;
if (t >= busy_until && t >= next_ns) {
const auto w0 = std::chrono::steady_clock::now();
const Stats before = tracker.stats;
const auto out = tracker.step(images, int64_t(t));
double ms = std::chrono::duration<double, std::milli>(std::chrono::steady_clock::now() - w0).count();
if (cost) { // repeatable: rounds of model calls at typical live costs, per thread
const int hands = tracker.stats.hand_calls - before.hand_calls, palms = tracker.stats.palm_calls - before.palm_calls;
ms = (2 + 10.0 * ((hands + threads - 1) / threads) + 18.0 * ((palms + threads - 1) / threads)) / slow;
}
busy_ms += ms;
busy_until = t + uint64_t(ms * slow * 1e6) + 3'000'000; // + the ring hand-off
next_ns = t + uint64_t((tracker.interval() - 0.005) * 1e9);
++processed;
last_out = out;
bool l = false, r = false;
for (const Hand *h : out) {
(h->pts[0][0] < 0 ? l : r) = true;
Track &tr = tracks[h->id];
auto palm = [](const V3 *p) { return (p[0] + p[5] + p[9] + p[13] + p[17]) * 0.2; };
const V3 raw = palm(h->pts), sm = palm(h->smooth);
++hand_updates, near_face += norm(sm) < 0.2;
if (tr.line && ts - tr.t[0] >= 0.1) tr.line = 0; // a gap: the line starts over
if (tr.line >= 2 && tr.t[0] - tr.t[1] > 1e-3) {
const double k = (ts - tr.t[0]) / (tr.t[0] - tr.t[1]);
jit_raw.push_back(norm(raw - tr.raw[0] - (tr.raw[0] - tr.raw[1]) * k) * 1000);
jit_sm.push_back(norm(sm - tr.sm[0] - (tr.sm[0] - tr.sm[1]) * k) * 1000);
}
tr.raw[1] = tr.raw[0], tr.sm[1] = tr.sm[0], tr.t[1] = tr.t[0];
tr.raw[0] = raw, tr.sm[0] = sm, tr.t[0] = ts;
++tr.line;
if (!tr.sets) tr.first = ts;
tr.last = ts, ++tr.sets, tr.left += h->pts[0][0] < 0;
if (tl)
std::fprintf(tl, "%.3f %d %s %d %+.3f %+.3f %+.3f\n", ts, h->id, h->pts[0][0] < 0 ? "L" : "R", h->nviews,
h->pts[0][0], h->pts[0][1], h->pts[0][2]);
}
if (tl && out.empty()) std::fprintf(tl, "%.3f -\n", ts);
if (tl)
for (const Seen &v : tracker.views_now())
std::fprintf(tl, "%.3f view %d %s presence %.2f roi %.0f %.0f %.0f %.3f set %d\n", ts, v.hand, v.cam.c_str(),
v.lm.presence, v.roi.center[0], v.roi.center[1], v.roi.size, v.roi.rotation, index);
left += l, right += r, both += l && r;
++hist[std::min<size_t>(out.size(), 2)];
}
if (oracle > 0 && nsets % oracle == 0) {
const Stats keep = tracker.stats;
const auto seen = tracker.exhaustive(images);
tracker.stats = keep;
bool l = false, r = false;
for (const Seen &s : seen) {
(s.wrist[0] < 0 ? l : r) = true;
++o_by_cam[s.cam + (s.wrist[0] < 0 ? " L" : " R")];
}
bool tl_ = false, tr_ = false;
for (const Hand *h : last_out) (h->pts[0][0] < 0 ? tl_ : tr_) = true;
++o_sets;
o_left += l, o_right += r;
o_left_hit += l && tl_, o_right_hit += r && tr_;
o_left_extra += !l && tl_, o_right_extra += !r && tr_;
if (tl && ((!l && tl_) || (!r && tr_))) std::fprintf(tl, "%.3f oracle-extra %s%s set %d\n", ts, !l && tl_ ? "L" : "", !r && tr_ ? "R" : "", index);
}
} while (in.next(cams, px));
if (tl) std::fclose(tl);
const double secs = nsets > 1 ? nsets / 30.0 : 0;
const Stats &s = tracker.stats;
std::printf("%s: %d sets (%.0f s), processed %d (%.1f/s), %.1f ms per step\n", dir.c_str(), nsets, secs, processed,
processed / std::max(secs, 1e-9), busy_ms / std::max(processed, 1));
std::printf("hands per processed set: 0 %.0f%%, 1 %.0f%%, 2 %.0f%%; a hand on the left %.0f%%, right %.0f%%, both %.0f%%\n",
100.0 * hist[0] / processed, 100.0 * hist[1] / processed, 100.0 * hist[2] / processed,
100.0 * left / processed, 100.0 * right / processed, 100.0 * both / processed);
std::vector<double> lens[2];
for (auto &[id, tr] : tracks) lens[tr.left * 2 > tr.sets ? 0 : 1].push_back(tr.last - tr.first);
for (int k = 0; k < 2; ++k) {
double total = 0;
for (double d : lens[k]) total += d;
std::printf("%s tracks: %zu, median %.1f s, total %.0f s\n", k ? "right" : "left ", lens[k].size(), median(lens[k]), total);
}
std::printf("views lost %d, handoff misses %d, dups %d, splits %d; hands new %d, merged %d, forgotten %d\n", s.lost,
s.handoff_miss, s.dups, s.splits, s.created, s.merged, s.forgotten);
std::printf("model calls: palm %d (%.1f/s), hand %d (%.1f/s)\n", s.palm_calls, s.palm_calls / std::max(secs, 1e-9),
s.hand_calls, s.hand_calls / std::max(secs, 1e-9));
{
std::vector<double> r, step;
std::map<int, double> prev;
for (auto &[id, x] : s.mono_ratio) {
r.push_back(x);
if (prev.count(id)) step.push_back(std::fabs(x - prev[id]));
prev[id] = x;
}
std::sort(r.begin(), r.end());
std::sort(step.begin(), step.end());
if (!r.empty())
std::printf("single-view distance / stereo: 10%% %.2f, median %.2f, 90%% %.2f; change between frames median %.3f, 90%% %.3f\n",
r[r.size() / 10], r[r.size() / 2], r[r.size() * 9 / 10], step[step.size() / 2], step[step.size() * 9 / 10]);
}
std::printf("palms within 20 cm of the eyes: %d of %d hand updates\n", near_face, hand_updates);
std::sort(jit_raw.begin(), jit_raw.end());
std::sort(jit_sm.begin(), jit_sm.end());
if (!jit_raw.empty())
std::printf("palm jitter (off a straight line through the last two updates): measured median %.1f mm, 90%% %.1f mm; "
"published median %.1f mm, 90%% %.1f mm\n", jit_raw[jit_raw.size() / 2], jit_raw[jit_raw.size() * 9 / 10],
jit_sm[jit_sm.size() / 2], jit_sm[jit_sm.size() * 9 / 10]);
if (o_sets) {
std::printf("oracle, %d sets: a left hand findable in %d, the tracker had it in %d (%.0f%%); right %d, had %d (%.0f%%)\n",
o_sets, o_left, o_left_hit, 100.0 * o_left_hit / std::max(o_left, 1), o_right, o_right_hit,
100.0 * o_right_hit / std::max(o_right, 1));
std::printf(" tracker had a hand the full search didn't find: left %d, right %d\n", o_left_extra, o_right_extra);
std::printf(" found by camera:");
for (auto &[k, n] : o_by_cam) std::printf(" %s %d", k.c_str(), n);
std::printf("\n");
}
return 0;
}