Checkpoint of the CUAS real-time work before the JCuda->JNA GPU-layer migration:
- OpticalFlow.buildSeries mode-0 curt_en fork: generate the merged-CUAS stack via
  CuasRanging.prepareFpixels() (GPU, explicit) then run the CUDA-free CuasDetectRT;
  coexists with the oracle (oracle gated off when curt_en).
- CuasDetectRT: file + in-memory(ImagePlus) entries via shared ingest(); -OFFSET gains
  an L2 "age" slice (5->6 ch), per-level noise scale, -LEV0 uniform naming, -OFFSET-<model> suffix.
- infer_server.py: L2 track-age (masked 5x5 max-pool, AGE_THR=0.2/AGE_K=0.5),
  per-level noise normalization (sqrt(2)^(L-3) default, Java-sent scale), nch + noise_scale
  + CMD_STATUS protocol additions; auto model-switch in CuasDnnRemote.ensureServer.
- cuasSynth + cuasNoise list SET keys (shared synth dir / inline per-level scales).
- CuasRanging.saveUasFlightLogCsv: per-frame UAS truth -> <name>-UAS_DATA.tsv (mode-0 only).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CLAUDE: -OFFSET in px (s-first, full-frame meta, s-gated NaN); curt_dnn_thresh viz-only (recurrent gets full field)

-OFFSET (remote) reordered to {s,Vx,Vy,dx,dy} (s first -> ImageJ auto-ranges on it); Vx,Vy converted cells->px/level-frame (/vel_decimate); full-frame ROI written to the file metadata (self-describes its extent); Vx,Vy,dx,dy NaN'd where s<curt_dnn_thresh (s kept). curt_dnn_thresh is now VISUALIZATION-ONLY: the local inferROI feeds Layer 2 the FULL field (sThresh=0) so the recurrent gets the weak sub-threshold signal it integrates (no premature threshold = the LReLU lesson); dialog label/tooltip/decl updated to say viz-only, do-not-use-for-computation.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Checkboxes were laid out with fill=HORIZONTAL, stretching the (text-less) JCheckBox across the half-width cell so its click target reached the scrollbar - a near-scrollbar miss-click silently toggled them (this flipped 'DNN remote' off mid-session). Anchor checkboxes WEST at natural size; the rest of the cell is now dead space. Applies to all tabbed dialogs.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

runDnnRemote requests each level's scenes in chunks (REQ=64) via CuasDnnRemote.inferBatch instead of per-scene; the DGX runs them continuously (production-representative ~100ms/scene full-res) and applies the ghostbuster on the GPU in decode, so BOTH the ROI 121-cell field and the full-frame -OFFSET {dx,dy,s,Vx,Vy} are ghostbusted (dropped the Java-side ghostbust). Validated: local vs remote on the same weighted9_pm_s model -> max |diff| ~1e-4 (ORT per-pixel vs PyTorch shift-and-stitch fp). Full-res ~100ms/scene is the oracle; RT would use the 1/4-res single forward (~4.4ms).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

runDnnRemote() decoded the raw softmax*s field but skipped dnnGhostbust, so the untrained corner-velocity sidelobes (which the local CPU path zeros when curt_dnn_vmax>0) survived as background noise (~0.06 vs ~5e-4 on the MAX-all-v layer). Apply dnnGhostbust to the ROI field, mirroring the local path -> verified: ImageJ subtract of -HYPER-RECT MAX slices vs the CPU (17_UAS_REFACTORED) is exact zero. (Full-frame -OFFSET s is not yet ghostbusted - Java lacks the full 121-field; a DGX-side decode ghostbuster would cover that.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

detectTargets() can offload the DNN front-end to the GB10 DGX (curt_dnn_remote): CuasDnnRemote uploads the LoG-conditioned stack once, the DGX builds the pyramid and runs full-res shift-and-stitch per (level,scene), returning the full-frame {dx,dy,s,Vx,Vy} (-OFFSET) + ROI 121-cell softmax*s (-RECT/-HYPER-RECT). Auto-launches the server if down (bundled cuas_dnn/ scripts, or curt_dnn_remote_srcdir local-repo override - mirrors the GPU-kernel default-vs-override). Synthetic targets are mixed into the upload stack so synth works on the remote path. 4 curt_dnn_remote_* dialog params, grouped with the model fields. Local CPU path unchanged (curt_dnn_remote=false) for Layer 2. Validated end-to-end; shift-and-stitch is fp64-exact vs per-pixel.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Dead after the 3d3/matched-filter removal (predecessor at tag cuas-layer1):
render3d3Hyperstack, get3d3Radius + the 3d3 kernel state (kernel3d3_rad, indx_*_3d3)
with its constructor param/setup/call-site; TemporalKernelGenerator.generateKernelMatched/Direct;
CuasDetectRT SUFFIX_CONV3D3, CONV3D3_MODES, bilinearFrame. No live callers; build OK.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Layer-1 is DNN-only; predecessor code preserved at git tag cuas-layer1.
Removed: 3d3 coarse-velocity path (curt_3d3_*, rleak1, convolve3D3LReLU);
C5P matched-filter / Bayes-posterior / whitening (curt_c5_matched/en/post/.../white*/from_pix);
LReLU rectification (curt_rleak0/pyr -> linear alpha=1, LoG kept);
distribution-vote (-GVOTE/-GHEADS, voteScatterGrid); curt_dnn_t8frac.
28 dialog params dropped; dialogQuestions/dialogAnswers re-verified aligned (44 widgets = 44 getNext). ~1841 lines removed.
Deprecated (kept): curt_stage2_model, MF-S S convention.
Kept: DNN front-end, recurrent, synth-test, SUBAVG, pyramid, LoG (linear). Build OK.

Runtime-untested (dialog) — verify a run; restore point: tag cuas-layer1.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- IntersceneMatchParameters: add curt_synth_bg_avg (synth quick-run: average N=2^LEV real
  frames per output frame and force a single level — restores the fast synthetic+background path).
- CuasDetectRT: that fast synth path (decimate-average to a short sequence + single level),
  honoring "Time ROI to" for the averaged-sequence length.
- CuasDnnInfer: MF-S head reads channel 0 as the raw matched-filter path-sum (clamp>=0, no sigmoid).

Validated: weighted model tracks the real UAS sub-pixel at LEV2/3; linear conditioning
(LReLU off) is dropout-free and matches the no-LReLU training. Checkpoint before Layer 2.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- CuasStage2Infer: ONNX runner for the learned Stage-2 vote-refine (full-field conv).
- CuasDetectRT: voteScatter (s^2-weighted bilinear splat at tail T=P-V*(N-1)) + reg-branch
  wiring -> saves -VOTE heatmap + -REFINED detection when curt_stage2_model is set;
  curt_c5_en gates the heavy C5P convolve (DNN-only fast path).
- IntersceneMatchParameters: curt_c5_en + curt_stage2_model params (all 6 sites each).
- CuasDnnInfer: setSessionLogLevel(ERROR) to suppress benign ORT VerifyOutputSizes warnings.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

parse_images_txt() now accepts both frame_NNN names and timestamp names
(e.g. 1763232419_474661.png). Introduces sort_key (float) so entries
sort correctly regardless of naming style. fno stays -1 for timestamp
images since no integer frame number is present.
Co-authored-by: Claude <claude@elphel.com>

Serial (USB/CH340G, 115200 baud) G-code driver for Creality Ender 3.
Provides tap(), swipe(), home(), calibrate() over /dev/ttyUSB0.
Part of the NanoKVM + Ender 3 physical phone-control project.
Co-authored-by: Claude <claude@elphel.com>

*-ims.corr-xml in the scene root dir always has an INS-fused GPS fix
(did_ins_2-lla, fallback did_gps1_pos-lla) giving the representative
position of the sequence for map location.

Changes:
- find_xml_files: also yields ims_path (*-ims.corr-xml from scene root)
- parse_lla(): reads lat/lon from INS or GPS1 LLA entry
- lat/lon added to metrics, console tables, and TSV output
Co-authored-by: Claude <claude@elphel.com>

*-elevations_histogram.csv (row 2, col 1) gives flight altitude above
ground level for each sequence.  Low AGL (<25-50 m) correlates with
algorithm failures in dense-canopy areas.

Changes:
- find_xml_files: also yields hist_path (*-elevations_histogram.csv)
- parse_agl(): reads AGL from second row, first column (tab-separated)
- agl_m added to sequence metrics dict and TSV output
- --min_agl filter: hide sequences below given AGL from console output
- AGL column shown in console tables
Co-authored-by: Claude <claude@elphel.com>

Finite-diff vel1 was dominated by single-frame pose glitches from
algorithm failures (low altitude / dense canopy), giving 600+ deg/s
for sequences where _dt shows ~10 deg/s.

Changes:
- parse _dt entries (algorithm's smooth polynomial derivatives) as
  primary velocity metric — unaffected by pose glitches
- keep finite-diff as secondary; glitch = fd_max / dt_max
  (~1 = clean data, >>1 = pose jumps / algorithm failure)
- --max_glitch filter hides suspect sequences from console output
  (TSV always contains all rows for Calc analysis)
- removed windowed finite-diff (vel5) — _dt makes it redundant
Co-authored-by: Claude <claude@elphel.com>

Scans all scene dirs under a linked data root, parses each
*-INTERFRAME.corr-xml (latest version subdir), and computes per-sequence
angular metrics for A/T/R axes:
  - range (max−min, deg)
  - vel1: max consecutive |Δangle|/Δt (deg/s) — fast, noise-sensitive
  - velW: max W-frame windowed |Δangle|/Δt (deg/s) — robust (default W=5)

A/T/R Cessna-nadir convention documented: R=heading, T=banking, A=pitch.
Sequence-length/altitude note documented: metrics are Δt-normalised.
Outputs ranked console tables and optional TSV.
Co-authored-by: Claude <claude@elphel.com>

- flat column names (no _x_y_z suffix, single comparison set)
- XML angles in degrees (not radians) — more useful for Calc plotting
- auto TSV when --out ends in .tsv, CSV otherwise
- column comment block documents layout in the source
Co-authored-by: Claude <claude@elphel.com>

Documents the ImageJ (1-based slice:timestamp), INTERFRAME.corr-xml
(timestamp_microseconds key), and COLMAP (0-based frame_NNNN) conventions
in the module docstring so any frame can be cross-referenced across tools.
Also clarifies the underscore→dot timestamp conversion in parse_interframe_xml.
Co-authored-by: Claude <claude@elphel.com>

Timestamps are unambiguous when using sub-sequences (e.g. 378 of 498
frames).  0-based sequential scene_idx was misleading.

- console top-N table: Scene column replaced with Timestamp
- CSV: scene_idx column removed, timestamp is the identity column
- parse_interframe_xml: drop scene_idx from returned dicts
Co-authored-by: Claude <claude@elphel.com>

*-egomotion.csv is a human-readable output file for Calc analysis, not
a machine input source.  *-INTERFRAME.corr-xml is the internal state
always present alongside processed data and is the correct programmatic
source for x/y/z/a/t/r poses.

- remove parse_egomotion_csv() and --ego argument
- --xml is now required
- remove IMS/PIMU col_specs (not in XML)
- rename CSV output columns: ego_* → xml_*, ers_*_rad → xml_*_rad
Co-authored-by: Claude <claude@elphel.com>

--ego (egomotion.csv) was required but is not always present (it is an
optional output file).  --xml (*-INTERFRAME.corr-xml) is always written
by the pipeline and contains the same x/y/z/a/t/r values.

Changes:
- add parse_interframe_xml(): sorts timestamps → same frame order as COLMAP
- --xml is the new preferred input; --ego is optional and overrides --xml
  (it additionally provides IMS/PIMU columns)
- at least one of --xml or --ego is required
- IMS/PIMU [SKIP] messages suppressed when using XML (NaN is expected)
- verified: XML and egomotion.csv produce identical results
Co-authored-by: Claude <claude@elphel.com>

CuasDnnInfer auto-detects reg from the ONNX output channels (6 = reg head
{det,Vx,Vy,logvar,dx,dy}, else grid) via getOutCh()/isReg(), and adds
inferROIReg() returning {S,Vx,Vy,sigma,dx,dy} per ROI pixel (S=sigmoid(det),
Vx/Vy model-bounded, sigma=exp(logvar/2)). CuasDetectRT branches: reg saves an
S-first {S,Vx,Vy,sigma} hyperstack (-VXYS) + -OFFSET (showArraysHyperstack,
channel x time x ROI), metadata-tagged; ghostbuster/grid renders/recurrent
skipped (no grid corners, |v|<=vmax baked in). Set curt_dnn_patch=32 for the
reg model; curt_dnn_vmax unused in reg mode.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

tagCuasImp() serializes every curt_* param (via IntersceneMatchParameters.setProperties,
so new params are auto-included) into the ImagePlus properties and calls
EyesisTiff.encodeProperiesToInfo -> XML in the "Info" field -> FileSaver writes it
to the TIFF ImageDescription. Applied to -DNN-RECT/-HYPER-RECT/-OFFSET. The python
reader (python3-imagej-tiff format) decodes it back, so outputs are self-describing;
ROI/model/patch/vmax/etc. restore without manual entry (supplements the filename tag).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The DNN save titles (-DNN-OFFSET / -HYPER-RECT / -RECT) now carry the ROI as
-ROI<x>_<y>_<w>_<h> (underscores to avoid scp/shell colon hazards) so the python
flight-log comparison can auto-restore the ROI per file instead of a hardcoded
constant - lets the real-UAS evaluation run hands-off across multiple ROIs.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The CONV2D images are saved BEFORE synth injection (real-only), so the actual
DNN input (LoG-conditioned real bg + injected synth) was no longer visible once
the mixing order moved to post-conditioning. Add a per-selected-level save of
dpixels_pyramid right after injection: getBaseName()+"-SYNTHMIX"+("B" if over
real bg)+("-SUBAVG" if avg-subtracted)+"-LEVn", full-frame (640x512), same
timestamps - what the network actually sees. Gated by save_LoG_pixels.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

inferROI's receptive-field patch was hardcoded P=24. Add CuasDnnInfer.setPatch
and new param curt_dnn_patch (default 24) so a wider-attention model (patch 32)
can be deployed: CuasDetectRT sets it on the inferer. The 32-patch model widens
off-center suppression reach (off_max 9->13 px) and removes the trajectory-alias
ghosts seen in the clean synthetic grid (field ~0.16 -> ~0.003). Must match the
loaded model's training patch.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The DNN velocity grid (radius vel_radius) extends past the trained velocity
disk, so untrained corner cells emit spurious sidelobes (ghosts) of non-trivial
strength (s up to ~0.09) that would confuse the recurrent layer. New
dnnGhostbust() zeros velocity cells with cell-radius > curt_dnn_vmax*vel_decimate,
and if a pixel's peak lands in the untrained corner it discards the whole
detection (field=0, s=0). Applied to the DNN field + offset before save and the
recurrent feed. New param curt_dnn_vmax (px/frame, default 1.4 = PM models'
training vmax_px; set to match the loaded model; <=0 disables).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

curt_synth_bg was fully parsed/persisted in IntersceneMatchParameters
but never read by the detector: the synthetic grid was always ADDED on
top of the LoG-conditioned real pyramid, so "Synthetic over real
background = false" had no effect (real background noise, and any NaN
border pixels, always present). Now when curt_synth_bg=false the level
is zeroed before the synthetic targets are injected -> truly clean,
noise-free targets. Logs which path was taken.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Synthetic mode previously force-overrode the ROI with the full-grid
SYNTH_ROI (320x320). showConvKernel5d expands each ROI pixel by
(vout_dim+1)*sub_dim per axis, so 320x320 blows the heap. Now synthetic
mode uses curt_save_select directly and falls back to SYNTH_ROI only
when no ROI is set, so the user can shrink it to a slice of the grid.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CuasDnnInfer reads N (temporal frame count) from the model's input
shape [B,N,H,W] and exposes getNFrames(); the constructor arg is now
only a fallback hint. CuasDetectRT uses dnn.getNFrames() instead of the
hardcoded 8, so an 8- vs 9-frame ONNX swaps purely by changing
curt_dnn_model (N follows automatically).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

- curt_subtract_avg (+ -SUBAVG filename tag): subtract the input per-pixel temporal
  mean before LoG. Removes the static treeline edge (and so its 8px tile-grid
  horizontal streak and the first-LReLU amplification of it); the moving target is
  not in the average, so it survives. Uses the whole sequence (not realtime; realtime
  would use a prior-run average).
- DNN compute-window: the timing ROI (curt_time_from/to) now gates the DNN inference
  loop (only in-window scenes are inferred), not just the saved output - fast iteration
  on a target's few seconds.
- DNN -> recurrent layer: feed the DNN field to runRecurrentLevel (per selected level,
  curt_recur_*). curt_dnn_recur_splat toggles feed-as-is vs bilinear splat of each
  pixel's velocity vector to its fractional (px+dx,py+dy) so neighbours reinforce in one
  sub-pixel bin (-SPLAT mark). curt_dnn_recur_scale (default 10) lifts the [0,1] field
  (peaks ~0.1) to the recurrent's tuned rs_min~1.0 scale. splatField() helper added.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CLAUDE: CUAS-RT overhaul - real-base synthetic-grid injection, level/time ROIs, 3d3 bypass, DNN offset/stride/window

Reworks the synthetic B-measurement and detector flow on the real pyramid base:

- Load: the real -CUAS-MERGED-CUAS stack is always the pyramid base (full-length
  levels); when curt_synth_src, the synthetic grid is loaded separately (scaled,
  NOT LoG'd) and injected TILED into each selected pyramid level AFTER the pure
  pyramid is built - same grid at every level, real noise averaged (sqrt(2)/level,
  printed). Both file picks use newestFile(). The old load-time mixing and the
  curt_synth_bg_avg (N=) decimation are retired.
- Level selection unified on curt_c5_levels for ALL heavy paths (3d3/conv5d,
  C5P-direct, DNN); cheap temporalAverage pyramid still builds up to the max
  selected level. Pyramid-depth guard stops building when a level would have <1
  frame (was NegativeArraySizeException on short sequences).
- Timing ROI curt_time_from/to (timestamp seconds; last-4-of-seconds shortcut,
  rollover-aware, 3-decimal fixed display) clips saved RECT/HYPER/DNN/C5P stacks
  and gates the DNN inference (compute-window, not just save).
- curt_3d3_en (default false) bypasses the 3d3 coarse-velocity path and its
  ~W*H*9 arrays. Full-velocities hyperstack render is guarded by a free-heap
  estimate (it is also an unfilled stub) and skipped if it would not fit.
- DNN: curt_dnn_stride (1 testing / 4 production 50%-overlap), and -DNN-OFFSET
  visualization (dx, dy, s over the ROI) from the offset channels.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

CuasRTUtils.getNumVout()/getSubDim() expose the 5D render dimensions for the
full-velocity-hyperstack memory estimate. CuasDnnInfer adds an inferROI(...,
double[][] offOut) overload that also returns the two offset channels
{dx, dy, s} per pixel (the DNN's per-pixel sub-pixel position estimate); the
existing 6-arg call delegates with offOut=null, so the field output is
unchanged. Both are prerequisites used by the CUAS-RT overhaul that follows.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Prepend velocity index 0 = max over all velocities (per scene, per pixel) to
showConvKernel5dHyperRect, and scale the display range to that slice instead
of the fixed IMP_MINMAX, so the small (~0.1) maxima are visible on open. It is
the first slice ImageJ shows - a quick map of non-empty pixels before scanning
scenes (Z) and velocities (T). Applies to both -HYPER-RECT and -DNN-HYPER-RECT.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

The bg_avg>1 path kept all n_synth frames and only added an averaged
background, so AVG4 and AVG8 produced the same frame count and 1/fps
timestamps. Now output frame i is the mean of bg_avg consecutive frames of
BOTH the scaled synthetic target AND the real background (non-overlapping
groups): the sequence is decimated by N (n_synth/N frames), the moving target
is motion-averaged like a real pyramid level, and each timestamp is the last
real frame of its group (N/fps increments).
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

When curt_dnn_model is set, run the ONNX DNN per-pixel inference over
curt_save_select across all available scenes (not just the first few),
saving -DNN-RECT / -DNN-HYPER-RECT. Slice labels carry timestamp + absolute
frame index for matching to real targets. New curt_dnn_thresh (6 sites in
IntersceneMatchParameters) zeroes the (Vx,Vy,s) field below confidence s -
display/FP-suppression only. inferROI tiles the ROI batch (CHUNK) to bound
ONNX Runtime activation memory.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

PyTorch-trained, ONNX-exported all-conv FCN (per-pixel Vx,Vy,s) run in Java via
ONNX Runtime 1.20.0 (CPU EP). CuasDnnInfer loads the model with a location resolver
(local path / scp user@host:path / http(s) -> ~/.cache/c5p_dnn/, fetching the
model.onnx + external-data .data pair) and runs a float[N][H][W] patch to raw
det/vel/off output. Verified bit-exact vs PyTorch (max abs diff 2.9e-6) via a fixed
test vector. New config param curt_dnn_model (empty default) selects the model,
mirroring the tile_processor_gpu kernel-source default/override scheme. CPU first
(.224 has no cuDNN; 82k net is microseconds/patch); GPU (CUDA/TensorRT EP) and the
CuasDetectRT integration are the next phase.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>