dgenerate-ultralytics-headless 8.3.239__py3-none-any.whl → 8.3.240__py3-none-any.whl

{dgenerate_ultralytics_headless-8.3.239.dist-info → dgenerate_ultralytics_headless-8.3.240.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dgenerate-ultralytics-headless
-Version: 8.3.239
+Version: 8.3.240
 Summary: Automatically built Ultralytics package with python-opencv-headless dependency instead of python-opencv
 Author-email: Glenn Jocher <glenn.jocher@ultralytics.com>, Jing Qiu <jing.qiu@ultralytics.com>
 Maintainer-email: Ultralytics <hello@ultralytics.com>
@@ -57,6 +57,7 @@ Provides-Extra: export
 Requires-Dist: numpy<2.0.0; extra == "export"
 Requires-Dist: onnx>=1.12.0; platform_system != "Darwin" and extra == "export"
 Requires-Dist: onnx<1.18.0,>=1.12.0; platform_system == "Darwin" and extra == "export"
+Requires-Dist: onnxslim>=0.1.80; extra == "export"
 Requires-Dist: coremltools>=9.0; (platform_system != "Windows" and python_version <= "3.13") and extra == "export"
 Requires-Dist: scikit-learn>=1.3.2; (platform_system != "Windows" and python_version <= "3.13") and extra == "export"
 Requires-Dist: openvino>=2024.0.0; extra == "export"

{dgenerate_ultralytics_headless-8.3.239.dist-info → dgenerate_ultralytics_headless-8.3.240.dist-info}/RECORD

@@ -1,4 +1,4 @@
-dgenerate_ultralytics_headless-8.3.239.dist-info/licenses/LICENSE,sha256=DZak_2itbUtvHzD3E7GNUYSRK6jdOJ-GqncQ2weavLA,34523
+dgenerate_ultralytics_headless-8.3.240.dist-info/licenses/LICENSE,sha256=DZak_2itbUtvHzD3E7GNUYSRK6jdOJ-GqncQ2weavLA,34523
 tests/__init__.py,sha256=bCox_hLdGRFYGLb2kd722VdNP2zEXNYNuLLYtqZSrbw,804
 tests/conftest.py,sha256=mOy9lGpNp7lk1hHl6_pVE0f9cU-72gnkoSm4TO-CNZU,2318
 tests/test_cli.py,sha256=GhIFHi-_WIJpDgoGNRi0DnjbfwP1wHbklBMnkCM-P_4,5464
@@ -8,7 +8,7 @@ tests/test_exports.py,sha256=8dxll33sqhlHeWplly5xxSdT-Nqw2cRRZWN3jZeRWWc,14155
 tests/test_integrations.py,sha256=6QgSh9n0J04RdUYz08VeVOnKmf4S5MDEQ0chzS7jo_c,6220
 tests/test_python.py,sha256=viMvRajIbDZdm64hRRg9i8qZ1sU9frwB69e56mxwEXk,29266
 tests/test_solutions.py,sha256=CIaphpmOXgz9AE9xcm1RWODKrwGfZLCc84IggGXArNM,14122
-ultralytics/__init__.py,sha256=Kkc0zFGRthNdbe2UViHpuowgEH5NVweklAPQWaMxVHg,1302
+ultralytics/__init__.py,sha256=O1bEyzXhjMQlfHguL1N3w9J7Xh5FIOC83E2vphoVoJw,1302
 ultralytics/py.typed,sha256=la67KBlbjXN-_-DfGNcdOcjYumVpKG_Tkw-8n5dnGB4,8
 ultralytics/assets/bus.jpg,sha256=wCAZxJecGR63Od3ZRERe9Aja1Weayrb9Ug751DS_vGM,137419
 ultralytics/assets/zidane.jpg,sha256=Ftc4aeMmen1O0A3o6GCDO9FlfBslLpTAw0gnetx7bts,50427
@@ -155,7 +155,7 @@ ultralytics/models/sam/amg.py,sha256=aYvJ7jQMkTR3X9KV7SHi3qP3yNchQggWNUurTRZwxQg
 ultralytics/models/sam/build.py,sha256=rEaFXA4R1nyutSonIenRKcuNtO1FgEojnkcayo0FTP4,12867
 ultralytics/models/sam/build_sam3.py,sha256=Gg_LiqNrCDTYaDWrob05vj-ln2AhkfMa5KkKhyk5wdE,11976
 ultralytics/models/sam/model.py,sha256=cOawDSkFqJPbt3455aTZ8tjaoWshFWFHQGGqxzsL_QQ,7372
-ultralytics/models/sam/predict.py,sha256=g-p88aNSeWql6bsjRb5FmjOXHGKvG0aJ9aIIlub7juU,203078
+ultralytics/models/sam/predict.py,sha256=rvUXBzNMvKlfjWP6UoKNm9GGpLcxJGsZGTkFq-7UOhI,203675
 ultralytics/models/sam/modules/__init__.py,sha256=lm6MckFYCPTbqIoX7w0s_daxdjNeBeKW6DXppv1-QUM,70
 ultralytics/models/sam/modules/blocks.py,sha256=ZU2aY4h6fmosj5pZ5EOEuO1O8Cl8UYeH11eOxkqCt8M,44570
 ultralytics/models/sam/modules/decoders.py,sha256=G4li37ahUe5rTTNTKibWMsAoz6G3R18rI8OPvfunVX8,25045
@@ -164,7 +164,7 @@ ultralytics/models/sam/modules/memory_attention.py,sha256=jFVWVbgDS7VXPqOL1e3gAz
 ultralytics/models/sam/modules/sam.py,sha256=-KV-1PZK39DTdSpR5DI3E8I6gGVLja3tMv1MH7Au_eA,61654
 ultralytics/models/sam/modules/tiny_encoder.py,sha256=RJQTHjfUe2N3cm1EZHXObJlKqVn10EnYJFla1mnWU_8,42065
 ultralytics/models/sam/modules/transformer.py,sha256=NmTuyxS9PNsg66tKY9_Q2af4I09VW5s8IbfswyTT3ao,14892
-ultralytics/models/sam/modules/utils.py,sha256=ztihxg0ssx0W-CKiqV-8KzB4og39TKnbmV3YO96ENPw,20770
+ultralytics/models/sam/modules/utils.py,sha256=Re09hcKe9LJpFzKHxpbwEmoMpPImnJaoNw7pe62_ui0,21129
 ultralytics/models/sam/sam3/__init__.py,sha256=aM4-KimnYgIFe-e5ctLT8e6k9PagvuvKFaHaagDZM7E,144
 ultralytics/models/sam/sam3/decoder.py,sha256=kXgPOjOh63ttJPFwMF90arK9AKZwPmhxOiexnPijiTE,22872
 ultralytics/models/sam/sam3/encoder.py,sha256=zP934Q_7B2I9UJvQY4Ghimu7Y8dQjBCGqUmSv5zbsq0,21794
@@ -269,7 +269,7 @@ ultralytics/utils/metrics.py,sha256=pP1LoKweL3Iga5dvCBVoqMSi5TwaaHERKKugDUloUs8,
 ultralytics/utils/nms.py,sha256=zv1rOzMF6WU8Kdk41VzNf1H1EMt_vZHcbDFbg3mnN2o,14248
 ultralytics/utils/ops.py,sha256=Se_Vb_smotVZ4g9gW4x56LpflAmbNDCEaeQz5BmUiD0,25637
 ultralytics/utils/patches.py,sha256=mD3slAMAhcezzP42_fOWmacNMU6zXB68Br4_EBCyIjs,7117
-ultralytics/utils/plotting.py,sha256=ivlhx4PZP80O-E8r4ruwxLYQEc6BLq5fmveKHgSi-EA,48219
+ultralytics/utils/plotting.py,sha256=bDOi3hmFaiJBuQyI3Ew-arliDNStuRIP7oM_Gq8C7Ac,48219
 ultralytics/utils/tal.py,sha256=w7oi6fp0NmL6hHh-yvCCX1cBuuB4JuX7w1wiR4_SMZs,20678
 ultralytics/utils/torch_utils.py,sha256=zOPUQlorTiEPSkqlSEPyaQhpmzmgOIKF7f3xJb0UjdQ,40268
 ultralytics/utils/tqdm.py,sha256=5PtGvRE9Xq8qugWqBSvZefAoFOnv3S0snETo5Z_ohNE,16185
@@ -291,8 +291,8 @@ ultralytics/utils/export/__init__.py,sha256=Cfh-PwVfTF_lwPp-Ss4wiX4z8Sm1XRPklsqd
 ultralytics/utils/export/engine.py,sha256=23-lC6dNsmz5vprSJzaN7UGNXrFlVedNcqhlOH_IXes,9956
 ultralytics/utils/export/imx.py,sha256=UHIq_PObOphIxctgSi0-5WaHvolHsHd3r5TTSjQSdgo,12860
 ultralytics/utils/export/tensorflow.py,sha256=igYzwbdblb9YgfV4Jgl5lMvynuVRcF51dAzI7j-BBI0,9966
-dgenerate_ultralytics_headless-8.3.239.dist-info/METADATA,sha256=PgCv3h3IXNlDfEVMUUSY2zWtTEV0HThO2cyZfavlRak,38747
-dgenerate_ultralytics_headless-8.3.239.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-dgenerate_ultralytics_headless-8.3.239.dist-info/entry_points.txt,sha256=YM_wiKyTe9yRrsEfqvYolNO5ngwfoL4-NwgKzc8_7sI,93
-dgenerate_ultralytics_headless-8.3.239.dist-info/top_level.txt,sha256=XP49TwiMw4QGsvTLSYiJhz1xF_k7ev5mQ8jJXaXi45Q,12
-dgenerate_ultralytics_headless-8.3.239.dist-info/RECORD,,
+dgenerate_ultralytics_headless-8.3.240.dist-info/METADATA,sha256=PT70UGSI51ivueJSk6BunjPAeZNEjMe9pr2HIf3JKKU,38798
+dgenerate_ultralytics_headless-8.3.240.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+dgenerate_ultralytics_headless-8.3.240.dist-info/entry_points.txt,sha256=YM_wiKyTe9yRrsEfqvYolNO5ngwfoL4-NwgKzc8_7sI,93
+dgenerate_ultralytics_headless-8.3.240.dist-info/top_level.txt,sha256=XP49TwiMw4QGsvTLSYiJhz1xF_k7ev5mQ8jJXaXi45Q,12
+dgenerate_ultralytics_headless-8.3.240.dist-info/RECORD,,

ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
-__version__ = "8.3.239"
+__version__ = "8.3.240"
 
 import importlib
 import os

ultralytics/models/sam/modules/utils.py

@@ -210,9 +210,14 @@ def apply_rotary_enc(
         # No keys to rotate, due to dropout
         return xq_out.type_as(xq).to(xq.device), xk
     # Repeat freqs along seq_len dim to match k seq_len
-    if repeat_freqs_k:
-        r = xk_.shape[-2] // xq_.shape[-2]
-        freqs_cis = freqs_cis.repeat(*([1] * (freqs_cis.ndim - 2)), r, 1)
+    if repeat_freqs_k and (r := xk_.shape[-2] // xq_.shape[-2]) > 1:
+        # MPS doesn't support repeat on complex tensors, decompose to real representation
+        if freqs_cis.device.type == "mps":
+            freqs_cis = torch.view_as_real(freqs_cis)
+            freqs_cis = freqs_cis.repeat(*([1] * (freqs_cis.ndim - 3)), r, 1, 1)
+            freqs_cis = torch.view_as_complex(freqs_cis.contiguous())
+        else:
+            freqs_cis = freqs_cis.repeat(*([1] * (freqs_cis.ndim - 2)), r, 1)
     xk_out = torch.view_as_real(xk_ * freqs_cis).flatten(3)
     return xq_out.type_as(xq).to(xq.device), xk_out.type_as(xk).to(xk.device)
 

ultralytics/models/sam/predict.py

@@ -878,6 +878,7 @@ class SAM2VideoPredictor(SAM2Predictor):
         self.clear_non_cond_mem_around_input = False
         self.clear_non_cond_mem_for_multi_obj = False
         self.callbacks["on_predict_start"].append(self.init_state)
+        self.clear_non_cond_mem = True  # Whether to clear non-conditioning memory periodically
 
     def get_model(self):
         """Retrieve and configure the model with binarization enabled.
@@ -952,6 +953,7 @@ class SAM2VideoPredictor(SAM2Predictor):
                 run_mem_encoder=True,
             )
             output_dict[storage_key][frame] = current_out
+            self._prune_non_cond_memory(frame)
         # Create slices of per-object outputs for subsequent interaction with each
         # individual object after tracking.
         self._add_output_per_object(frame, current_out, storage_key)
@@ -1830,6 +1832,25 @@ class SAM2VideoPredictor(SAM2Predictor):
         inference_state["frames_already_tracked"].clear()
         inference_state["first_ann_frame_idx"] = None
 
+    def _prune_non_cond_memory(self, frame_idx, inference_state=None):
+        """Prune old non-conditioning frames to bound memory usage."""
+        if not self.clear_non_cond_mem:
+            return
+        inference_state = inference_state or self.inference_state
+
+        # Determine window size
+        min_frame = frame_idx - self.model.num_maskmem * self.model.memory_temporal_stride_for_eval
+        output_dict = inference_state["output_dict"]
+
+        # Prune global non_cond_frame_outputs
+        for f in [k for k in output_dict["non_cond_frame_outputs"] if k < min_frame]:
+            output_dict["non_cond_frame_outputs"].pop(f, None)
+
+        # Prune per-object non_cond_frame_outputs
+        for obj_output_dict in inference_state.get("output_dict_per_obj", {}).values():
+            for f in [k for k in obj_output_dict["non_cond_frame_outputs"] if k < min_frame]:
+                obj_output_dict["non_cond_frame_outputs"].pop(f, None)
+
 
 class SAM2DynamicInteractivePredictor(SAM2Predictor):
     """SAM2DynamicInteractivePredictor extends SAM2Predictor to support dynamic interactions with video frames or a
@@ -2423,6 +2444,7 @@ class SAM3VideoPredictor(SAM2VideoPredictor, SAM3Predictor):
                 inference_state=inference_state,
             )
             output_dict[storage_key][frame] = current_out
+            self._prune_non_cond_memory(frame, inference_state=inference_state)
         # Create slices of per-object outputs for subsequent interaction with each
         # individual object after tracking.
         self._add_output_per_object(frame, current_out, storage_key, inference_state=inference_state)
@@ -2475,14 +2497,12 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
         hotstart_delay=0,
         hotstart_unmatch_thresh=3,
         hotstart_dup_thresh=3,
-        # Whether to suppress masks only within hotstart. If False, we can suppress masks even if they start before hotstart period.
-        suppress_unmatched_only_within_hotstart=True,
-        init_trk_keep_alive=0,
-        max_trk_keep_alive=8,
+        init_trk_keep_alive=30,
+        max_trk_keep_alive=30,
         min_trk_keep_alive=-4,
         # Threshold for suppressing overlapping objects based on recent occlusion
         suppress_overlapping_based_on_recent_occlusion_threshold=0.0,
-        decrease_trk_keep_alive_for_empty_masklets=False,
+        decrease_trk_keep_alive_for_empty_masklets=True,
         o2o_matching_masklets_enable=False,  # Enable hungarian matching to match existing masklets
         suppress_det_close_to_boundary=False,
         fill_hole_area=16,
@@ -2513,7 +2533,6 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
         self.hotstart_delay = hotstart_delay
         self.hotstart_unmatch_thresh = hotstart_unmatch_thresh
         self.hotstart_dup_thresh = hotstart_dup_thresh
-        self.suppress_unmatched_only_within_hotstart = suppress_unmatched_only_within_hotstart
         self.init_trk_keep_alive = init_trk_keep_alive
         self.max_trk_keep_alive = max_trk_keep_alive
         self.min_trk_keep_alive = min_trk_keep_alive
@@ -2638,7 +2657,7 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
                 ) > 0
 
         # names = getattr(self.model, "names", [str(i) for i in range(pred_scores.shape[0])])
-        names = dict(enumerate(str(i) for i in range(pred_masks.shape[0])))
+        names = dict(enumerate(str(i) for i in range(pred_boxes.shape[0])))
         results = []
         for masks, boxes, orig_img, img_path in zip([pred_masks], [pred_boxes], orig_imgs, self.batch[0]):
             results.append(Results(orig_img, path=img_path, names=names, masks=masks, boxes=boxes))
@@ -2689,7 +2708,6 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
         metadata = tracker_metadata_new["metadata"]
         removed_obj_ids = metadata["removed_obj_ids"]
         out["removed_obj_ids"] = removed_obj_ids
-        out["suppressed_obj_ids"] = metadata["suppressed_obj_ids"][frame_idx]
         out["frame_stats"] = frame_stats
         if self.masklet_confirmation_enable:
             status = metadata["masklet_confirmation"]["status"]
@@ -3597,7 +3615,6 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
         overlap_pair_to_frame_inds = metadata["overlap_pair_to_frame_inds"]
         # removed_obj_ids: object IDs that are suppressed via hot-start
         removed_obj_ids = metadata["removed_obj_ids"]
-        suppressed_obj_ids = metadata["suppressed_obj_ids"][frame_idx]
 
         obj_ids_newly_removed = set()  # object IDs to be newly removed on this frame
         hotstart_diff = frame_idx - self.hotstart_delay if not reverse else frame_idx + self.hotstart_delay
@@ -3647,12 +3664,12 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
                     )
             if (
                 trk_keep_alive[obj_id] <= 0  # Object has not been matched for too long
-                and not self.suppress_unmatched_only_within_hotstart
                 and obj_id not in removed_obj_ids
                 and obj_id not in obj_ids_newly_removed
             ):
-                LOGGER.debug(f"Suppressing object {obj_id} at frame {frame_idx}, due to being unmatched")
-                suppressed_obj_ids.add(obj_id)
+                LOGGER.debug(f"Removing object {obj_id} at frame {frame_idx}, due to being unmatched")
+                # directly removed the object instead of suppressing it
+                obj_ids_newly_removed.add(obj_id)
 
         # Step 3: removed tracks that overlaps with another track for `hotstart_dup_thresh` frames
         # a) find overlaps tracks -- we consider overlap if they match to the same detection
@@ -3831,8 +3848,6 @@ class SAM3VideoSemanticPredictor(SAM3SemanticPredictor):
             "trk_keep_alive": defaultdict(int),  # This is used only for object suppression not for removal
             "overlap_pair_to_frame_inds": defaultdict(list),
             "removed_obj_ids": set(),
-            # frame_idx --> set of objects with suppressed outputs, but still continue to be tracked
-            "suppressed_obj_ids": defaultdict(set),
         }
         if self.masklet_confirmation_enable:
             # all the following are np.ndarray with the same shape as `obj_ids_all_gpu`

ultralytics/utils/plotting.py

@@ -24,17 +24,6 @@ class Colors:
     This class provides methods to work with the Ultralytics color palette, including converting hex color codes to RGB
     values and accessing predefined color schemes for object detection and pose estimation.
 
-    Attributes:
-        palette (list[tuple]): List of RGB color tuples for general use.
-        n (int): The number of colors in the palette.
-        pose_palette (np.ndarray): A specific color palette array for pose estimation with dtype np.uint8.
-
-    Examples:
-        >>> from ultralytics.utils.plotting import Colors
-        >>> colors = Colors()
-        >>> colors(5, True)  # Returns BGR format: (221, 111, 255)
-        >>> colors(5, False)  # Returns RGB format: (255, 111, 221)
-
     ## Ultralytics Color Palette
 
     | Index | Color                                                             | HEX       | RGB               |
@@ -89,6 +78,17 @@ class Colors:
 
         For Ultralytics brand colors see [https://www.ultralytics.com/brand](https://www.ultralytics.com/brand).
         Please use the official Ultralytics colors for all marketing materials.
+
+    Attributes:
+        palette (list[tuple]): List of RGB color tuples for general use.
+        n (int): The number of colors in the palette.
+        pose_palette (np.ndarray): A specific color palette array for pose estimation with dtype np.uint8.
+
+    Examples:
+        >>> from ultralytics.utils.plotting import Colors
+        >>> colors = Colors()
+        >>> colors(5, True)  # Returns BGR format: (221, 111, 255)
+        >>> colors(5, False)  # Returns RGB format: (255, 111, 221)
     """
 
     def __init__(self):