L'électron rare
67302e7ca2
Three perf optimizations stacked on top of the distributed pipeline: 1. multihmr_server.py ported to pyobjc CoreML.framework direct (drops the ~30 ms coremltools.MLModel.predict overhead). Compiles the mlpackage to .mlmodelc on load, then predicts via MLDictionaryFeatureProvider + MLMultiArray ctypes memcpy. Toggle MULTIHMR_SERVER_BACKEND=coremltools for fallback. setup script installs pyobjc-framework-CoreML on the macm1 venv. 2. MediaPipe Holistic now uses GPU Metal delegate on M5. Required wrapping camera frames as SRGBA (4-channel) instead of SRGB -- the Metal CVPixelBuffer path rejects 3-channel formats. Bench M5 standalone : pose 6.7 -> 2.9 ms, face 4.0 -> 1.0 ms, hand 6.1 -> 3.2 ms. Frees ~10 ms CPU per frame for OSC + rigger. 3. Remote client queues bumped maxsize 1 -> 2/3 (in/out) to absorb jitter without stalling capture.
84 lines
2.8 KiB
Bash
Executable File
84 lines
2.8 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
# Push the Multi-HMR mlpackage + server to macm1 (M1 Max, 32-core GPU)
|
|
# and launch the inference server in the background.
|
|
#
|
|
# Prereqs on macm1 :
|
|
# * passwordless ssh (Tailscale alias 'macm1' or LAN)
|
|
# * uv installed
|
|
# * Python 3.12 available via uv (uv pulls it)
|
|
#
|
|
# Usage:
|
|
# ./scripts/setup_remote_macm1.sh
|
|
# MACM1_HOST=clems@192.168.0.175 ./scripts/setup_remote_macm1.sh
|
|
set -euo pipefail
|
|
|
|
HOST="${MACM1_HOST:-macm1}"
|
|
PORT="${MULTIHMR_SERVER_PORT:-57140}"
|
|
MLPACKAGE_LOCAL="${MLPACKAGE_LOCAL:-$HOME/.cache/av-live-multihmr/multihmr_full_672_s.mlpackage}"
|
|
REMOTE_TMP="/tmp/av-live-multihmr"
|
|
REMOTE_VENV="/tmp/av-live-multihmr/venv"
|
|
|
|
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
|
|
|
echo "==> Target host : $HOST"
|
|
echo "==> mlpackage : $MLPACKAGE_LOCAL"
|
|
|
|
if [ ! -d "$MLPACKAGE_LOCAL" ]; then
|
|
echo "ERROR: mlpackage missing at $MLPACKAGE_LOCAL" >&2
|
|
exit 1
|
|
fi
|
|
|
|
echo "==> Creating remote tmp dir"
|
|
ssh "$HOST" "mkdir -p $REMOTE_TMP"
|
|
|
|
echo "==> rsync mlpackage (~70 MB, may take a moment first time)"
|
|
rsync -a --delete \
|
|
"$MLPACKAGE_LOCAL/" \
|
|
"$HOST:$REMOTE_TMP/multihmr_full_672_s.mlpackage/"
|
|
|
|
echo "==> rsync multihmr_server.py"
|
|
rsync -a "$SCRIPT_DIR/multihmr_server.py" \
|
|
"$HOST:$REMOTE_TMP/multihmr_server.py"
|
|
|
|
echo "==> Provision Python 3.12 venv with uv (idempotent)"
|
|
ssh "$HOST" "bash -lc 'set -e
|
|
if [ ! -x $REMOTE_VENV/bin/python ]; then
|
|
uv venv --python 3.12 $REMOTE_VENV --quiet
|
|
fi
|
|
uv pip install --python $REMOTE_VENV/bin/python --quiet \
|
|
coremltools numpy opencv-python-headless \
|
|
pyobjc-core pyobjc-framework-Cocoa pyobjc-framework-CoreML
|
|
'"
|
|
|
|
echo "==> Killing any stale server on :$PORT"
|
|
ssh "$HOST" "bash -lc 'pkill -f multihmr_server.py 2>/dev/null || true; sleep 0.3'"
|
|
|
|
echo "==> Launching server (background)"
|
|
ssh "$HOST" "bash -lc 'cd $REMOTE_TMP && \
|
|
MULTIHMR_SERVER_PORT=$PORT \
|
|
nohup $REMOTE_VENV/bin/python multihmr_server.py \
|
|
--mlpackage $REMOTE_TMP/multihmr_full_672_s.mlpackage \
|
|
--port $PORT \
|
|
>> $REMOTE_TMP/server.log 2>&1 &
|
|
echo \$! > $REMOTE_TMP/server.pid
|
|
disown || true'"
|
|
|
|
echo "==> Waiting for server to be ready"
|
|
REMOTE_ADDR=$(ssh "$HOST" 'echo $SSH_CONNECTION' | awk '{print $3}')
|
|
# Fallback to host alias if SSH_CONNECTION trick fails.
|
|
if [ -z "${REMOTE_ADDR:-}" ]; then REMOTE_ADDR="$HOST"; fi
|
|
|
|
for i in $(seq 1 30); do
|
|
if ssh "$HOST" "bash -lc 'nc -z 127.0.0.1 $PORT 2>/dev/null'"; then
|
|
echo "==> Server up on $HOST:$PORT (probed via localhost on host)"
|
|
echo "==> Reachable from this Mac at $REMOTE_ADDR:$PORT"
|
|
ssh "$HOST" "tail -n 20 $REMOTE_TMP/server.log" || true
|
|
exit 0
|
|
fi
|
|
sleep 1
|
|
done
|
|
|
|
echo "ERROR: server did not come up within 30s. Last log lines:" >&2
|
|
ssh "$HOST" "tail -n 60 $REMOTE_TMP/server.log" || true
|
|
exit 1
|