Fix degenerate detection: bypass multi-image patching during probe

trellis2/pipelines/trellis2_image_to_3d.py

@@ -404,6 +404,9 @@ class Trellis2ImageTo3DPipeline(Pipeline):
         """
         Run 2 probe steps to detect degenerate texture flow trajectory.
 
+        Uses single-image inference (bypassing multi-image patching) so
+        the threshold stays calibrated regardless of fusion mode.
+
         Returns True if the trajectory is degenerate (will produce black texture).
         """
         steps = sampler_params.get('steps', 12)
@@ -412,16 +415,32 @@ class Trellis2ImageTo3DPipeline(Pipeline):
         t_seq = rescale_t * t_seq / (1 + (rescale_t - 1) * t_seq)
         t_seq = t_seq.tolist()
 
-        sample = noise
-        for step_idx in range(2):
-            t, t_prev = t_seq[step_idx], t_seq[step_idx + 1]
-            out = self.tex_slat_sampler.sample_once(
-                flow_model, sample, t, t_prev,
-                concat_cond=concat_cond,
-                **{k: v for k, v in cond.items()},
-                **{k: v for k, v in sampler_params.items() if k not in ('steps', 'rescale_t')},
-            )
-            sample = out.pred_x_prev
+        # Use single-image cond for probe (threshold was calibrated on single-image)
+        probe_cond = {}
+        for k, v in cond.items():
+            probe_cond[k] = v[:1] if torch.is_tensor(v) and v.ndim >= 1 else v
+
+        # Bypass multi-image patching during probe
+        sampler = self.tex_slat_sampler
+        patched = hasattr(sampler, '_old_inference_model')
+        if patched:
+            patched_fn = sampler._inference_model
+            sampler._inference_model = sampler._old_inference_model
+
+        try:
+            sample = noise
+            for step_idx in range(2):
+                t, t_prev = t_seq[step_idx], t_seq[step_idx + 1]
+                out = sampler.sample_once(
+                    flow_model, sample, t, t_prev,
+                    concat_cond=concat_cond,
+                    **{k: v for k, v in probe_cond.items()},
+                    **{k: v for k, v in sampler_params.items() if k not in ('steps', 'rescale_t')},
+                )
+                sample = out.pred_x_prev
+        finally:
+            if patched:
+                sampler._inference_model = patched_fn
 
         x0_std = out.pred_x_0.feats.std().item()
         return x0_std > threshold