CLAUDE: in-process native DNN backend (CuasDnnLocal) via JNA - no server

Piece 3 of the native-JNA DNN path. Adds a local backend that runs the SAME
L1+L2 inference as CuasDnnRemote but in-process via LibTorch (libtpdnn.so / JNA),
so the CUAS pipeline runs without the DGX or any Python server:
  - CuasDnnBackend  : shared interface (upload/getNFrames/inferBatch->BatchResult/close)
  - TpDnnJna        : JNA Library binding libtpdnn.so's C-ABI
  - CuasDnnLocal    : wraps it; reads N/P/vr/l2_ch from each model's bundled .meta.json
                      (single source of truth), float[][]<->float[], builds BatchResult
  - CuasDnnRemote   : now implements CuasDnnBackend (signatures unchanged)
  - CuasDetectRT    : DNN path gate now fires on (curt_dnn_remote || curt_dnn_local);
                      backend = local? CuasDnnLocal : CuasDnnRemote; ensureServer skipped
                      when local; local-CPU-ORT gate also excludes curt_dnn_local (no
                      double-run). runDnnRemote loop unchanged.
  - IntersceneMatchParameters: curt_dnn_local (flag) + curt_dnn_local_dir (model dir
                      override; empty = bundled /cuas_dnn resource) + GUI labels/persist.

Validated: full Java->JNA->libtpdnn vs the Python-server oracle = EXACT
(offset5=0.0, roi=0.0, nch=6). mvn -DskipTests package OK.

Runtime: -Djna.library.path=<dir with libtpdnn.so>; libtpdnn.so finds libtorch via
its rpath. Model resolution mirrors CuasDnnRemote's bundled-vs-override scheme.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

src/main/java/com/elphel/imagej/cuas/rt/CuasDetectRT.java

@@ -748,9 +748,12 @@ public class CuasDetectRT {
 		final boolean synth_bg = clt_parameters.imp.curt_synth_bg;
 		final int n_synth = synth ? synth_pixels.length : 0;
 		try { // auto-launch the DGX server if not already running (deploy bundled/override scripts, ssh-start, poll) // By Claude on 06/20/2026
+			if (!clt_parameters.imp.curt_dnn_local) // local native backend has no server to launch. By Claude on 06/27/2026
 					CuasDnnRemote.ensureServer(clt_parameters.imp.curt_dnn_remote_host, clt_parameters.imp.curt_dnn_remote_model, l2model, clt_parameters.imp.curt_dnn_remote_srcdir);
 		} catch (Exception e) { System.out.println("runDnnRemote(): server auto-launch failed: "+e); }
-		try (CuasDnnRemote remote = new CuasDnnRemote(clt_parameters.imp.curt_dnn_remote_host)) {
+		try (CuasDnnBackend remote = clt_parameters.imp.curt_dnn_local
+					? new CuasDnnLocal(clt_parameters.imp.curt_dnn_local_dir, clt_parameters.imp.curt_dnn_remote_model, l2model)
+					: new CuasDnnRemote(clt_parameters.imp.curt_dnn_remote_host)) {
 			// Build the upload array: LoG-conditioned real (optionally synth_bg_avg-decimated upstream), optionally
 			// with synthetic targets mixed in (tiled synth[t % n_synth]; clean = zero bg first) - matching the local
 			// per-level injection (CuasDetectRT.java synth block) but PRE-pyramid; the DGX then averages it into levels.
@@ -1150,7 +1153,7 @@ public class CuasDetectRT {
    		// same shape as the C5P output, saved as -DNN-RECT / -DNN-HYPER-RECT. N (temporal
    		// depth, 8 or 9) is read from the loaded ONNX. Reuses curt_c5_levels for level gating. Recurrent feed is the next step
    		// (the field is [0,1]-scaled, unlike the C5P matched-filter response - needs rescale).
-   		if (!clt_parameters.imp.curt_dnn_model.isEmpty() && (curt_save_select != null) && !clt_parameters.imp.curt_dnn_remote) { // local CPU path (remote = DGX block below) // By Claude on 06/13/2026, remote guard 06/20/2026
+   		if (!clt_parameters.imp.curt_dnn_model.isEmpty() && (curt_save_select != null) && !clt_parameters.imp.curt_dnn_remote && !clt_parameters.imp.curt_dnn_local) { // local CPU path (remote=DGX / curt_dnn_local=native LibTorch blocks below) // By Claude on 06/13/2026, remote guard 06/20/2026, local guard 06/27/2026
    			final int vr_dnn = clt_parameters.imp.curt_vel_radius;                            // By Claude on 06/13/2026
    			final int dnn_stride = Math.max(1, clt_parameters.imp.curt_dnn_stride);           // 1 = every slice (testing), 4 = production 50% overlap // By Claude on 06/14/2026
    			try {                                                                             // By Claude on 06/13/2026
@@ -1317,7 +1320,7 @@ public class CuasDetectRT {
    			}                                                                                 // By Claude on 06/13/2026
    		}                                                                                     // By Claude on 06/13/2026
    		// Remote DGX inference path (curt_dnn_remote): upload conditioned stack, DGX builds pyramid + infers. // By Claude on 06/20/2026
-   		if ((curt_save_select != null) && clt_parameters.imp.curt_dnn_remote) { // model lives on the DGX (run dir) - no local curt_dnn_model needed // By Claude on 06/20/2026
+   		if ((curt_save_select != null) && (clt_parameters.imp.curt_dnn_remote || clt_parameters.imp.curt_dnn_local)) { // DGX server OR in-process native LibTorch (curt_dnn_local). By Claude on 06/27/2026
    			runDnnRemote(clt_parameters, cuasRTUtils, dpixels_log, ts_pyramid, pyramid_levels, c5_levels, curt_save_select, title_conv5d); // By Claude on 06/20/2026
    		}
    		System.out.println(now()+" detectTargets(): done"); // By Claude on 06/11/2026

src/main/java/com/elphel/imagej/cuas/rt/CuasDnnBackend.java

+package com.elphel.imagej.cuas.rt;
+
+import java.awt.Rectangle;
+
+/**
+ * DNN inference backend for the CUAS L1+L2 pipeline. Implemented by {@link CuasDnnRemote} (TCP to the
+ * Python infer_server, e.g. on the DGX) and {@link CuasDnnLocal} (in-process native LibTorch via JNA /
+ * libtpdnn.so). CuasDetectRT.runDnnRemote() talks to this interface so the two are drop-in swappable;
+ * the result type {@link CuasDnnRemote.BatchResult} is shared. By Claude on 06/27/2026.
+ */
+public interface CuasDnnBackend extends AutoCloseable {
+	/** Upload the conditioned stack frames[T][H*W] (row-major) once; the backend builds the temporal
+	 *  pyramid. Returns the per-level frame counts (and sets {@link #getNFrames()}). */
+	int [] upload(float [][] frames, int H, int W) throws Exception;
+
+	/** Temporal depth N of the loaded model (window length), known after {@link #upload}. */
+	int getNFrames();
+
+	/** Infer `count` scenes of a level in one call (newest_s = start + s*stride). rmaxCells>0 enables the
+	 *  on-GPU ghostbuster. l2Enable/l2Reset drive Layer-2; noiseScale is the per-level L1-input scale. */
+	CuasDnnRemote.BatchResult inferBatch(int level, int start, int count, int stride, Rectangle roi,
+			double rmaxCells, boolean l2Enable, boolean l2Reset, double noiseScale) throws Exception;
+
+	@Override
+	void close();
+}

src/main/java/com/elphel/imagej/cuas/rt/CuasDnnLocal.java

+package com.elphel.imagej.cuas.rt;
+
+import java.awt.Rectangle;
+import java.io.File;
+import java.io.InputStream;
+import java.nio.file.Files;
+import java.nio.file.Path;
+import java.nio.file.StandardCopyOption;
+import java.util.regex.Matcher;
+import java.util.regex.Pattern;
+
+import com.sun.jna.Native;
+import com.sun.jna.Pointer;
+
+/**
+ * In-process native DNN backend: runs the SAME L1+L2 inference as {@link CuasDnnRemote}, but via
+ * LibTorch through JNA (libtpdnn.so, tile_processor_gpu/jna/tp_dnn.cpp) instead of TCP to a Python
+ * server. No DGX / no server process. Each model's params (N/P/vr/out_ch, L2 ch_hidden) come from its
+ * bundled {@code <model>.ts.pt.meta.json} sidecar (single source of truth). By Claude on 06/27/2026.
+ *
+ * Model resolution mirrors CuasDnnRemote's bundled-vs-override convention:
+ *   - localDir set  -> &lt;localDir&gt;/&lt;model&gt;/model.ts.pt (+ .meta.json)   [dev/override]
+ *   - localDir empty -> bundled jar resource /cuas_dnn/&lt;basename&gt;/model.ts.pt extracted to temp
+ *
+ * Requires {@code -Djna.library.path=<dir with libtpdnn.so>}; libtpdnn.so finds libtorch via its rpath
+ * (set by build_dnn.sh) or LD_LIBRARY_PATH.
+ */
+public class CuasDnnLocal implements CuasDnnBackend {
+	private static TpDnnJna LIB = null;
+	private static synchronized TpDnnJna lib() {
+		if (LIB == null) LIB = Native.load("tpdnn", TpDnnJna.class);
+		return LIB;
+	}
+
+	private final TpDnnJna   lib;
+	private final Pointer    ctx;
+	private final int        N, P, vr, nvel;
+	private int              nFrames = 0;
+	private int              H = 0, W = 0;
+	private File             tmpDir = null;     // holds extracted bundled models (deleted on close)
+
+	/** @param localDir  override dir holding &lt;model&gt;/model.ts.pt, or empty/null to use bundled resources.
+	 *  @param l1Model   L1 run name (e.g. "runs/weighted9_pm_s").
+	 *  @param l2Model   L2 run name (e.g. "runs/mexhat_gaps_boost40"), or empty/null for L1-only. */
+	public CuasDnnLocal(String localDir, String l1Model, String l2Model) throws Exception {
+		this.lib = lib();
+		String [] l1 = resolveModel(localDir, l1Model, ".ts.pt");      // {tsPath, metaJson}
+		this.N  = metaInt(l1[1], "N", 9);
+		this.P  = metaInt(l1[1], "P", 24);
+		this.vr = metaInt(l1[1], "vr", 5);
+		this.nvel = (2 * vr + 1) * (2 * vr + 1);
+		String l2ts = "";
+		int l2ch = 24;
+		if ((l2Model != null) && !l2Model.isEmpty()) {
+			String [] l2 = resolveModel(localDir, l2Model, ".l2.ts.pt");
+			l2ts = l2[0];
+			l2ch = metaInt(l2[1], "ch_hidden", 24);
+		}
+		this.ctx = lib.tpdnn_init(l1[0], l2ts, N, P, vr, l2ch);
+		if (ctx == null) throw new Exception("tpdnn_init failed (l1=" + l1[0] + ", l2=" + l2ts + ")");
+		System.out.println("CuasDnnLocal: native LibTorch backend up (L1=" + l1Model
+				+ ", L2=" + ((l2Model == null || l2Model.isEmpty()) ? "off" : l2Model)
+				+ "; N=" + N + " P=" + P + " vr=" + vr + " l2_ch=" + l2ch + ")");
+	}
+
+	@Override
+	public int [] upload(float [][] frames, int H, int W) throws Exception {
+		this.H = H; this.W = W;
+		int T = frames.length, hw = H * W;
+		float [] flat = new float [T * hw];                            // [T,H,W] row-major
+		for (int t = 0; t < T; t++) System.arraycopy(frames[t], 0, flat, t * hw, hw);
+		lib.tpdnn_upload(ctx, flat, T, H, W);
+		this.nFrames = N;
+		int nl = lib.tpdnn_num_levels(ctx);
+		int [] counts = new int [nl];
+		for (int i = 0; i < nl; i++) counts[i] = lib.tpdnn_level_frames(ctx, i);
+		return counts;
+	}
+
+	@Override
+	public int getNFrames() { return nFrames; }
+
+	@Override
+	public CuasDnnRemote.BatchResult inferBatch(int level, int start, int count, int stride, Rectangle roi,
+			double rmaxCells, boolean l2Enable, boolean l2Reset, double noiseScale) throws Exception {
+		int hw = H * W, rn = roi.width * roi.height;
+		int nchAlloc = l2Enable ? 6 : 5;                              // native returns 6 with L2 (+age), else 5
+		float [] o5  = new float [count * nchAlloc * hw];
+		float [] rfb = new float [count * rn * nvel];
+		long t0 = System.nanoTime();
+		int nch = lib.tpdnn_infer(ctx, level, start, count, stride,
+				roi.x, roi.y, roi.width, roi.height, rmaxCells,
+				l2Enable ? 1 : 0, l2Reset ? 1 : 0, noiseScale, o5, rfb);
+		double ms = (System.nanoTime() - t0) / 1.0e6;
+
+		CuasDnnRemote.BatchResult r = new CuasDnnRemote.BatchResult();
+		r.gpuMs = ms; r.H = H; r.W = W; r.count = count; r.nch = nch; r.nvel = nvel;
+		r.rh = roi.height; r.rw = roi.width;
+		r.offset5 = new float [count][nch][hw];                       // o5 layout [count][nch][hw]
+		for (int s = 0; s < count; s++)
+			for (int c = 0; c < nch; c++)
+				System.arraycopy(o5, (s * nch + c) * hw, r.offset5[s][c], 0, hw);
+		r.roiField = new float [count][rn][nvel];                     // rfb layout [count][rn][nvel]
+		for (int s = 0; s < count; s++)
+			for (int p = 0; p < rn; p++)
+				System.arraycopy(rfb, (s * rn + p) * nvel, r.roiField[s][p], 0, nvel);
+		return r;
+	}
+
+	@Override
+	public void close() {
+		try { if (ctx != null) lib.tpdnn_free(ctx); } catch (Exception e) { /* ignore */ }
+		if (tmpDir != null) { try { deleteRec(tmpDir); } catch (Exception e) { /* ignore */ } }
+	}
+
+	// ---- model resolution + tiny meta parser ----
+
+	/** Returns {tsPath, metaJsonContent} for the model. localDir override, else bundled resource->temp. */
+	private String [] resolveModel(String localDir, String model, String suffix) throws Exception {
+		String fileName = "model" + suffix;                          // model.ts.pt / model.l2.ts.pt
+		if ((localDir != null) && !localDir.isEmpty()) {
+			File ts   = new File(new File(localDir, model), fileName);
+			File meta = new File(ts.getPath() + ".meta.json");
+			if (!ts.exists()) throw new Exception("local model not found: " + ts);
+			String mj = meta.exists() ? new String(Files.readAllBytes(meta.toPath())) : "";
+			return new String [] { ts.getPath(), mj };
+		}
+		// bundled: /cuas_dnn/<basename(model)>/model.ts.pt (+ .meta.json) -> extract to a temp dir
+		String base = model.substring(model.lastIndexOf('/') + 1);
+		if (tmpDir == null) tmpDir = Files.createTempDirectory("cuas_dnn_").toFile();
+		String res = "/cuas_dnn/" + base + "/" + fileName;
+		Path tsTmp = extractResource(res, base + suffix);
+		String mj = "";
+		try (InputStream is = CuasDnnLocal.class.getResourceAsStream(res + ".meta.json")) {
+			if (is != null) mj = new String(is.readAllBytes());
+		}
+		return new String [] { tsTmp.toString(), mj };
+	}
+
+	private Path extractResource(String res, String tmpName) throws Exception {
+		try (InputStream is = CuasDnnLocal.class.getResourceAsStream(res)) {
+			if (is == null) throw new Exception("bundled resource not found: " + res);
+			Path out = new File(tmpDir, tmpName).toPath();
+			Files.copy(is, out, StandardCopyOption.REPLACE_EXISTING);
+			return out;
+		}
+	}
+
+	private static int metaInt(String json, String key, int dflt) {
+		if (json == null) return dflt;
+		Matcher m = Pattern.compile("\"" + key + "\"\\s*:\\s*(-?\\d+)").matcher(json);
+		return m.find() ? Integer.parseInt(m.group(1)) : dflt;
+	}
+
+	private static void deleteRec(File f) {
+		File [] kids = f.listFiles();
+		if (kids != null) for (File k : kids) deleteRec(k);
+		f.delete();
+	}
+}

src/main/java/com/elphel/imagej/cuas/rt/CuasDnnRemote.java

@@ -28,7 +28,7 @@ import java.nio.file.Files;
  *   READBACK: cmd=3, int level, frame -> reply: int H,W, H*W float32 (one conditioned pyramid frame)
  *   BYE     : cmd=0
  */
-public class CuasDnnRemote implements AutoCloseable {
+public class CuasDnnRemote implements CuasDnnBackend {
 	private static final int CMD_BYE = 0, CMD_UPLOAD = 1, CMD_INFER = 2, CMD_READBACK = 3, CMD_STATUS = 4;
 	private final Socket sock;
 	private final DataInputStream  in;

src/main/java/com/elphel/imagej/cuas/rt/TpDnnJna.java

+package com.elphel.imagej.cuas.rt;
+
+import com.sun.jna.Library;
+import com.sun.jna.Pointer;
+
+/**
+ * JNA boundary to the native LibTorch DNN inference library (libtpdnn.so, tile_processor_gpu/jna/tp_dnn.cpp).
+ * Mirrors the C-ABI: load TorchScript L1(+L2), build the temporal pyramid from an uploaded stack, and run
+ * shift-and-stitch + decode (+ L2 recurrence/age) per level. Loaded by base name "tpdnn" via
+ * {@code Native.load} with {@code -Djna.library.path=<dir with libtpdnn.so>}. By Claude on 06/27/2026.
+ */
+public interface TpDnnJna extends Library {
+	/** Load TorchScript L1 (+optional L2) onto CUDA. l2_path empty/NULL -> L1-only. Returns an opaque ctx. */
+	Pointer tpdnn_init(String l1_path, String l2_path, int N, int P, int vr, int l2_ch);
+
+	/** Build the temporal pyramid from host log [T,H,W] (row-major float32). */
+	void tpdnn_upload(Pointer ctx, float[] log, int T, int H, int W);
+
+	int tpdnn_num_levels(Pointer ctx);
+	int tpdnn_level_frames(Pointer ctx, int level);
+
+	/** One INFER over `count` scenes of `level`. Writes offset5 into out_o5 [count*nch*H*W] and roi into
+	 *  out_roi [count*rh*rw*nvel] (native float order). Returns nch (5 L1-only, 6 with L2:+age). */
+	int tpdnn_infer(Pointer ctx, int level, int start, int count, int stride,
+			int rx, int ry, int rw, int rh, double rmax,
+			int l2_enable, int l2_reset, double noise_scale,
+			float[] out_o5, float[] out_roi);
+
+	void tpdnn_free(Pointer ctx);
+}

src/main/java/com/elphel/imagej/tileprocessor/IntersceneMatchParameters.java

@@ -1168,6 +1168,8 @@ min_str_neib_fpn	0.35
 	public String   curt_dnn_remote_host = "192.168.0.62:5577"; // DGX inference server host:port for curt_dnn_remote (see attic/imagej-elphel-internal/c5p_dnn/infer_server.py) // By Claude on 06/20/2026
 	public String   curt_dnn_remote_model = "runs/weighted9_pm_s"; // DGX-side run dir (with model.pt) the auto-launched server loads (passed as RUN=) // By Claude on 06/20/2026
 	public String   curt_dnn_remote_srcdir = ""; // server-scripts override dir: empty = bundled jar resource (cuas_dnn/), set = local dir - same default-vs-override scheme as the GPU kernels (cuda_project_directory) // By Claude on 06/20/2026
+	public boolean  curt_dnn_local =       false; // run the DNN front-end in-process via native LibTorch (CuasDnnLocal/libtpdnn.so) instead of the remote Python server - identical L1+L2 result, no DGX/server. Overrides curt_dnn_remote when set. // By Claude on 06/27/2026
+	public String   curt_dnn_local_dir =   ""; // override dir holding <model>/model.ts.pt (+.meta.json) for curt_dnn_local: empty = bundled jar resource (cuas_dnn/) - same default-vs-override scheme as the kernels. // By Claude on 06/27/2026
 	public boolean  curt_dnn_l2 =          false; // run the trained Layer-2 (track-before-detect ConvGRU) on the DGX after L1: -OFFSET then carries L2 {det,Vx,Vy} (L1's full-res, NON-ghostbusted field fed in, recurrence over the scene/time axis). Off = L1 offset5 as before (re-run with this off to inspect L1). Requires curt_dnn_remote. // By Claude on 06/22/2026
 	public String   curt_dnn_l2_model =    "runs/l2_v1"; // DGX-side Layer-2 run dir (with model.pt) the auto-launched server loads (passed as RUN2=) // By Claude on 06/22/2026
 	public boolean  curt_dnn_recur_splat = false; // when feeding the DNN field to the recurrent layer: false = feed per-pixel field as-is; true = splat each pixel's velocity vector to its fractional offset (px+dx,py+dy) so neighbours reinforce in one sub-pixel bin // By Claude on 06/14/2026
@@ -3480,6 +3482,10 @@ min_str_neib_fpn	0.35
 				"DGX-side run directory (containing model.pt) the auto-launched server loads, e.g. runs/weighted9_pm_s.");
 		gd.addStringField ("DNN remote server src (empty=bundled)",  this.curt_dnn_remote_srcdir, 40, // By Claude on 06/20/2026
 				"Override dir for the DGX server scripts (infer_server.py / run_infer_server.sh): empty = bundled jar resource (cuas_dnn/); set = local dir. Same default-vs-override (bundled resource vs local repo) scheme as the GPU kernels - bundled is the working version, refresh it after server-script dev.");
+		gd.addCheckbox    ("DNN local (in-process LibTorch)",        this.curt_dnn_local, // By Claude on 06/27/2026
+				"Run the DNN front-end in-process via native LibTorch (libtpdnn.so / JNA) instead of the remote Python server - identical L1+L2 result, no DGX/server. When set, overrides 'DNN remote'. Needs -Djna.library.path pointing at libtpdnn.so.");
+		gd.addStringField ("DNN local model dir (empty=bundled)",    this.curt_dnn_local_dir, 40, // By Claude on 06/27/2026
+				"Override dir holding <model>/model.ts.pt (+ .meta.json) for 'DNN local': empty = bundled jar resource (cuas_dnn/). Same default-vs-override scheme as the kernels. Uses the same model names as the remote fields.");
 		gd.addCheckbox    ("DNN Layer-2 (run on DGX)",               this.curt_dnn_l2, // By Claude on 06/22/2026
 				"Run the trained Layer-2 track-before-detect ConvGRU on the DGX after L1 (requires 'DNN remote'). -OFFSET then shows L2 {det,Vx,Vy} (L1's full-res non-ghostbusted field fed in, recurrence over time), titled -DNN-L2-. Uncheck to re-run the old L1 way."); // By Claude on 06/22/2026
 		gd.addStringField ("DNN Layer-2 model (DGX run dir)",        this.curt_dnn_l2_model, 24, // By Claude on 06/22/2026
@@ -5032,6 +5038,8 @@ min_str_neib_fpn	0.35
 		this.curt_dnn_remote_host =     gd.getNextString().trim(); // By Claude on 06/20/2026
 		this.curt_dnn_remote_model =    gd.getNextString().trim(); // By Claude on 06/20/2026
 		this.curt_dnn_remote_srcdir =   gd.getNextString().trim(); // By Claude on 06/20/2026
+		this.curt_dnn_local =           gd.getNextBoolean();     // By Claude on 06/27/2026
+		this.curt_dnn_local_dir =       gd.getNextString().trim(); // By Claude on 06/27/2026
 		this.curt_dnn_l2 =              gd.getNextBoolean();     // By Claude on 06/22/2026
 		this.curt_dnn_l2_model =        gd.getNextString().trim(); // By Claude on 06/22/2026
 		this.curt_dnn_thresh =          gd.getNextNumber();      // By Claude on 06/13/2026
@@ -6401,6 +6409,8 @@ min_str_neib_fpn	0.35
 		properties.setProperty(prefix+"curt_dnn_remote_host", this.curt_dnn_remote_host);   // String // By Claude on 06/20/2026
 		properties.setProperty(prefix+"curt_dnn_remote_model", this.curt_dnn_remote_model); // String // By Claude on 06/20/2026
 		properties.setProperty(prefix+"curt_dnn_remote_srcdir", this.curt_dnn_remote_srcdir); // String // By Claude on 06/20/2026
+		properties.setProperty(prefix+"curt_dnn_local",       this.curt_dnn_local+"");      // boolean // By Claude on 06/27/2026
+		properties.setProperty(prefix+"curt_dnn_local_dir",   this.curt_dnn_local_dir);     // String // By Claude on 06/27/2026
 		properties.setProperty(prefix+"curt_dnn_l2",          this.curt_dnn_l2+"");         // boolean // By Claude on 06/22/2026
 		properties.setProperty(prefix+"curt_dnn_l2_model",    this.curt_dnn_l2_model);      // String // By Claude on 06/22/2026
 		properties.setProperty(prefix+"curt_synth_src",       this.curt_synth_src+"");      // boolean // By Claude on 06/11/2026
@@ -6798,6 +6808,8 @@ min_str_neib_fpn	0.35
 		if (properties.getProperty(prefix+"curt_dnn_remote_host")!=null) this.curt_dnn_remote_host=(String) properties.getProperty(prefix+"curt_dnn_remote_host"); // By Claude on 06/20/2026
 		if (properties.getProperty(prefix+"curt_dnn_remote_model")!=null)  this.curt_dnn_remote_model=(String) properties.getProperty(prefix+"curt_dnn_remote_model"); // By Claude on 06/20/2026
 		if (properties.getProperty(prefix+"curt_dnn_remote_srcdir")!=null) this.curt_dnn_remote_srcdir=(String) properties.getProperty(prefix+"curt_dnn_remote_srcdir"); // By Claude on 06/20/2026
+		if (properties.getProperty(prefix+"curt_dnn_local")!=null)       this.curt_dnn_local=Boolean.parseBoolean(properties.getProperty(prefix+"curt_dnn_local")); // By Claude on 06/27/2026
+		if (properties.getProperty(prefix+"curt_dnn_local_dir")!=null)   this.curt_dnn_local_dir=(String) properties.getProperty(prefix+"curt_dnn_local_dir"); // By Claude on 06/27/2026
 		if (properties.getProperty(prefix+"curt_dnn_l2")!=null)         this.curt_dnn_l2=Boolean.parseBoolean(properties.getProperty(prefix+"curt_dnn_l2")); // By Claude on 06/22/2026
 		if (properties.getProperty(prefix+"curt_dnn_l2_model")!=null)   this.curt_dnn_l2_model=(String) properties.getProperty(prefix+"curt_dnn_l2_model"); // By Claude on 06/22/2026
 
@@ -9077,6 +9089,8 @@ min_str_neib_fpn	0.35
 		imp.curt_dnn_remote_host =  this.curt_dnn_remote_host; // By Claude on 06/20/2026
 		imp.curt_dnn_remote_model = this.curt_dnn_remote_model; // By Claude on 06/20/2026
 		imp.curt_dnn_remote_srcdir = this.curt_dnn_remote_srcdir; // By Claude on 06/20/2026
+		imp.curt_dnn_local =       this.curt_dnn_local;     // By Claude on 06/27/2026
+		imp.curt_dnn_local_dir =   this.curt_dnn_local_dir; // By Claude on 06/27/2026
 		imp.curt_dnn_l2 =           this.curt_dnn_l2;        // By Claude on 06/22/2026
 		imp.curt_dnn_l2_model =     this.curt_dnn_l2_model;  // By Claude on 06/22/2026
 		imp.curt_synth_src =        this.curt_synth_src;     // By Claude on 06/11/2026