dgenerate-ultralytics-headless 8.3.180__tar.gz → 8.3.181__tar.gz

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dgenerate-ultralytics-headless
-Version: 8.3.180
+Version: 8.3.181
 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>

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/dgenerate_ultralytics_headless.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dgenerate-ultralytics-headless
-Version: 8.3.180
+Version: 8.3.181
 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>

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/__init__.py

@@ -1,6 +1,6 @@
 # Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license
 
-__version__ = "8.3.180"
+__version__ = "8.3.181"
 
 import os
 

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/data/loaders.py

@@ -355,9 +355,10 @@ class LoadImagesAndVideos:
             channels (int): Number of image channels (1 for grayscale, 3 for RGB).
         """
         parent = None
-        if isinstance(path, str) and Path(path).suffix == ".txt":  # *.txt file with img/vid/dir on each line
-            parent = Path(path).parent
-            path = Path(path).read_text().splitlines()  # list of sources
+        if isinstance(path, str) and Path(path).suffix in {".txt", ".csv"}:  # txt/csv file with source paths
+            parent, content = Path(path).parent, Path(path).read_text()
+            path = content.splitlines() if Path(path).suffix == ".txt" else content.split(",")  # list of sources
+            path = [p.strip() for p in path]
         files = []
         for p in sorted(path) if isinstance(path, (list, tuple)) else [path]:
             a = str(Path(p).absolute())  # do not use .resolve() https://github.com/ultralytics/ultralytics/issues/2912

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/data/utils.py

@@ -219,9 +219,7 @@ def verify_image_label(args: Tuple) -> List:
                 assert lb.min() >= -0.01, f"negative class labels {lb[lb < -0.01]}"
 
                 # All labels
-                if single_cls:
-                    lb[:, 0] = 0
-                max_cls = lb[:, 0].max()  # max label count
+                max_cls = 0 if single_cls else lb[:, 0].max()  # max label count
                 assert max_cls < num_cls, (
                     f"Label class {int(max_cls)} exceeds dataset class count {num_cls}. "
                     f"Possible class labels are 0-{num_cls - 1}"

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/sam/modules/blocks.py

@@ -3,7 +3,7 @@
 import copy
 import math
 from functools import partial
-from typing import Any, Optional, Tuple, Type, Union
+from typing import Optional, Tuple, Type, Union
 
 import numpy as np
 import torch
@@ -856,8 +856,11 @@ class PositionEmbeddingRandom(nn.Module):
     def forward(self, size: Tuple[int, int]) -> torch.Tensor:
         """Generate positional encoding for a grid using random spatial frequencies."""
         h, w = size
-        device: Any = self.positional_encoding_gaussian_matrix.device
-        grid = torch.ones((h, w), device=device, dtype=torch.float32)
+        grid = torch.ones(
+            (h, w),
+            device=self.positional_encoding_gaussian_matrix.device,
+            dtype=self.positional_encoding_gaussian_matrix.dtype,
+        )
         y_embed = grid.cumsum(dim=0) - 0.5
         x_embed = grid.cumsum(dim=1) - 0.5
         y_embed = y_embed / h
@@ -871,7 +874,7 @@ class PositionEmbeddingRandom(nn.Module):
         coords = coords_input.clone()
         coords[:, :, 0] = coords[:, :, 0] / image_size[1]
         coords[:, :, 1] = coords[:, :, 1] / image_size[0]
-        return self._pe_encoding(coords.to(torch.float))  # B x N x C
+        return self._pe_encoding(coords)  # B x N x C
 
 
 class Block(nn.Module):

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/sam/modules/decoders.py

@@ -423,7 +423,7 @@ class SAM2MaskDecoder(nn.Module):
 
         # Upscale mask embeddings and predict masks using the mask tokens
         src = src.transpose(1, 2).view(b, c, h, w)
-        if not self.use_high_res_features:
+        if not self.use_high_res_features or high_res_features is None:
             upscaled_embedding = self.output_upscaling(src)
         else:
             dc1, ln1, act1, dc2, act2 = self.output_upscaling

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/sam/modules/encoders.py

@@ -258,8 +258,8 @@ class PromptEncoder(nn.Module):
         """Embed point prompts by applying positional encoding and label-specific embeddings."""
         points = points + 0.5  # Shift to center of pixel
         if pad:
-            padding_point = torch.zeros((points.shape[0], 1, 2), device=points.device)
-            padding_label = -torch.ones((labels.shape[0], 1), device=labels.device)
+            padding_point = torch.zeros((points.shape[0], 1, 2), dtype=points.dtype, device=points.device)
+            padding_label = -torch.ones((labels.shape[0], 1), dtype=labels.dtype, device=labels.device)
             points = torch.cat([points, padding_point], dim=1)
             labels = torch.cat([labels, padding_label], dim=1)
         point_embedding = self.pe_layer.forward_with_coords(points, self.input_image_size)
@@ -300,10 +300,6 @@ class PromptEncoder(nn.Module):
         else:
             return 1
 
-    def _get_device(self) -> torch.device:
-        """Return the device of the first point embedding's weight tensor."""
-        return self.point_embeddings[0].weight.device
-
     def forward(
         self,
         points: Optional[Tuple[torch.Tensor, torch.Tensor]],
@@ -334,7 +330,11 @@ class PromptEncoder(nn.Module):
             torch.Size([1, 7, 256]) torch.Size([1, 256, 64, 64])
         """
         bs = self._get_batch_size(points, boxes, masks)
-        sparse_embeddings = torch.empty((bs, 0, self.embed_dim), device=self._get_device())
+        sparse_embeddings = torch.empty(
+            (bs, 0, self.embed_dim),
+            dtype=self.point_embeddings[0].weight.dtype,
+            device=self.point_embeddings[0].weight.device,
+        )
         if points is not None:
             coords, labels = points
             point_embeddings = self._embed_points(coords, labels, pad=(boxes is None))
@@ -637,7 +637,7 @@ class FpnNeck(nn.Module):
             lateral_features = self.convs[n - i](x)
             if i in self.fpn_top_down_levels and prev_features is not None:
                 top_down_features = F.interpolate(
-                    prev_features.to(dtype=torch.float32),
+                    prev_features.to(dtype=x.dtype),
                     scale_factor=2.0,
                     mode=self.fpn_interp_model,
                     align_corners=(None if self.fpn_interp_model == "nearest" else False),

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/sam/modules/sam.py

@@ -488,7 +488,7 @@ class SAM2Model(torch.nn.Module):
             assert sam_point_coords.size(0) == B and sam_point_labels.size(0) == B
         else:
             # If no points are provide, pad with an empty point (with label -1)
-            sam_point_coords = torch.zeros(B, 1, 2, device=device)
+            sam_point_coords = torch.zeros(B, 1, 2, device=device, dtype=backbone_features.dtype)
             sam_point_labels = -torch.ones(B, 1, dtype=torch.int32, device=device)
 
         # b) Handle mask prompts
@@ -533,7 +533,6 @@ class SAM2Model(torch.nn.Module):
 
         # convert masks from possibly bfloat16 (or float16) to float32
         # (older PyTorch versions before 2.1 don't support `interpolate` on bf16)
-        low_res_multimasks = low_res_multimasks.float()
         high_res_multimasks = F.interpolate(
             low_res_multimasks,
             size=(self.image_size, self.image_size),
@@ -560,12 +559,11 @@ class SAM2Model(torch.nn.Module):
             if self.soft_no_obj_ptr:
                 lambda_is_obj_appearing = object_score_logits.sigmoid()
             else:
-                lambda_is_obj_appearing = is_obj_appearing.float()
+                lambda_is_obj_appearing = is_obj_appearing.to(obj_ptr.dtype)
 
             if self.fixed_no_obj_ptr:
                 obj_ptr = lambda_is_obj_appearing * obj_ptr
             obj_ptr = obj_ptr + (1 - lambda_is_obj_appearing) * self.no_obj_ptr
-
         return (
             low_res_multimasks,
             high_res_multimasks,
@@ -769,7 +767,7 @@ class SAM2Model(torch.nn.Module):
                     if self.add_tpos_enc_to_obj_ptrs:
                         t_diff_max = max_obj_ptrs_in_encoder - 1
                         tpos_dim = C if self.proj_tpos_enc_in_obj_ptrs else self.mem_dim
-                        obj_pos = torch.tensor(pos_list, device=device)
+                        obj_pos = torch.tensor(pos_list, device=device, dtype=current_vision_feats[-1].dtype)
                         obj_pos = get_1d_sine_pe(obj_pos / t_diff_max, dim=tpos_dim)
                         obj_pos = self.obj_ptr_tpos_proj(obj_pos)
                         obj_pos = obj_pos.unsqueeze(1).expand(-1, B, self.mem_dim)
@@ -834,7 +832,7 @@ class SAM2Model(torch.nn.Module):
         # scale the raw mask logits with a temperature before applying sigmoid
         binarize = self.binarize_mask_from_pts_for_mem_enc and is_mask_from_pts
         if binarize and not self.training:
-            mask_for_mem = (pred_masks_high_res > 0).float()
+            mask_for_mem = (pred_masks_high_res > 0).to(pix_feat.dtype)
         else:
             # apply sigmoid on the raw mask logits to turn them into range (0, 1)
             mask_for_mem = torch.sigmoid(pred_masks_high_res)
@@ -927,11 +925,10 @@ class SAM2Model(torch.nn.Module):
     ):
         """Run memory encoder on predicted mask to encode it into a new memory feature for future frames."""
         if run_mem_encoder and self.num_maskmem > 0:
-            high_res_masks_for_mem_enc = high_res_masks
             maskmem_features, maskmem_pos_enc = self._encode_new_memory(
                 current_vision_feats=current_vision_feats,
                 feat_sizes=feat_sizes,
-                pred_masks_high_res=high_res_masks_for_mem_enc,
+                pred_masks_high_res=high_res_masks,
                 object_score_logits=object_score_logits,
                 is_mask_from_pts=(point_inputs is not None),
             )

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/sam/modules/utils.py

@@ -78,7 +78,7 @@ def get_1d_sine_pe(pos_inds: torch.Tensor, dim: int, temperature: float = 10000)
         torch.Size([4, 128])
     """
     pe_dim = dim // 2
-    dim_t = torch.arange(pe_dim, dtype=torch.float32, device=pos_inds.device)
+    dim_t = torch.arange(pe_dim, dtype=pos_inds.dtype, device=pos_inds.device)
     dim_t = temperature ** (2 * (dim_t // 2) / pe_dim)
 
     pos_embed = pos_inds.unsqueeze(-1) / dim_t

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/sam/predict.py

@@ -132,9 +132,9 @@ class Predictor(BasePredictor):
             im = torch.from_numpy(im)
 
         im = im.to(self.device)
-        im = im.half() if self.model.fp16 else im.float()
         if not_tensor:
             im = (im - self.mean) / self.std
+        im = im.half() if self.model.fp16 else im.float()
         return im
 
     def pre_transform(self, im):
@@ -251,7 +251,6 @@ class Predictor(BasePredictor):
             labels (np.ndarray | List[int] | None): Point prompt labels with shape (N,). 1 = foreground, 0 = background.
             masks (List[np.ndarray] | np.ndarray | None): Masks for the objects, where each mask is a 2D array.
             multimask_output (bool): Flag to return multiple masks for ambiguous prompts.
-            img_idx (int): Index of the image in the batch to process.
 
         Returns:
             pred_masks (torch.Tensor): Output masks with shape (C, H, W), where C is the number of generated masks.
@@ -298,7 +297,7 @@ class Predictor(BasePredictor):
         r = 1.0 if self.segment_all else min(dst_shape[0] / src_shape[0], dst_shape[1] / src_shape[1])
         # Transform input prompts
         if points is not None:
-            points = torch.as_tensor(points, dtype=torch.float32, device=self.device)
+            points = torch.as_tensor(points, dtype=self.torch_dtype, device=self.device)
             points = points[None] if points.ndim == 1 else points
             # Assuming labels are all positive if users don't pass labels.
             if labels is None:
@@ -312,11 +311,11 @@ class Predictor(BasePredictor):
                 # (N, 2) --> (N, 1, 2), (N, ) --> (N, 1)
                 points, labels = points[:, None, :], labels[:, None]
         if bboxes is not None:
-            bboxes = torch.as_tensor(bboxes, dtype=torch.float32, device=self.device)
+            bboxes = torch.as_tensor(bboxes, dtype=self.torch_dtype, device=self.device)
             bboxes = bboxes[None] if bboxes.ndim == 1 else bboxes
             bboxes *= r
         if masks is not None:
-            masks = torch.as_tensor(masks, dtype=torch.float32, device=self.device).unsqueeze(1)
+            masks = torch.as_tensor(masks, dtype=self.torch_dtype, device=self.device).unsqueeze(1)
         return bboxes, points, labels, masks
 
     def generate(
@@ -450,7 +449,8 @@ class Predictor(BasePredictor):
         if model is None:
             model = self.get_model()
         model.eval()
-        self.model = model.to(device)
+        model = model.to(device)
+        self.model = model.half() if self.args.half else model.float()
         self.device = device
         self.mean = torch.tensor([123.675, 116.28, 103.53]).view(-1, 1, 1).to(device)
         self.std = torch.tensor([58.395, 57.12, 57.375]).view(-1, 1, 1).to(device)
@@ -459,8 +459,9 @@ class Predictor(BasePredictor):
         self.model.pt = False
         self.model.triton = False
         self.model.stride = 32
-        self.model.fp16 = False
+        self.model.fp16 = self.args.half
         self.done_warmup = True
+        self.torch_dtype = torch.float16 if self.model.fp16 else torch.float32
 
     def get_model(self):
         """Retrieve or build the Segment Anything Model (SAM) for image segmentation tasks."""
@@ -832,9 +833,10 @@ class SAM2Predictor(Predictor):
         Perform inference on image features using the SAM2 model.
 
         Args:
-            features (Dict[str, Any]): Extracted image information from the SAM2 model image encoder, it's a dictionary including:
-                image_embed (torch.Tensor): Image embedding with shape (B, C, H, W).
-                high_res_feats (List[torch.Tensor]): List of high-resolution feature maps from the backbone, each with shape (B, C, H, W).
+            features (torch.Tensor | Dict[str, Any]): Extracted image features with shape (B, C, H, W) from the SAM2 model image encoder, it
+                could also be a dictionary including:
+                - image_embed (torch.Tensor): Image embedding with shape (B, C, H, W).
+                - high_res_feats (List[torch.Tensor]): List of high-resolution feature maps from the backbone, each with shape (B, C, H, W).
             points (np.ndarray | List[List[float]] | None): Object location points with shape (N, 2), in pixels.
             labels (np.ndarray | List[int] | None): Point prompt labels with shape (N,). 1 = foreground, 0 = background.
             masks (List[np.ndarray] | np.ndarray | None): Masks for the objects, where each mask is a 2D array.
@@ -853,9 +855,12 @@ class SAM2Predictor(Predictor):
         )
         # Predict masks
         batched_mode = points is not None and points[0].shape[0] > 1  # multi object prediction
-        high_res_features = [feat_level[img_idx].unsqueeze(0) for feat_level in features["high_res_feats"]]
+        high_res_features = None
+        if isinstance(features, dict):
+            high_res_features = [feat_level[img_idx].unsqueeze(0) for feat_level in features["high_res_feats"]]
+            features = features["image_embed"][[img_idx]]
         pred_masks, pred_scores, _, _ = self.model.sam_mask_decoder(
-            image_embeddings=features["image_embed"][img_idx].unsqueeze(0),
+            image_embeddings=features,
             image_pe=self.model.sam_prompt_encoder.get_dense_pe(),
             sparse_prompt_embeddings=sparse_embeddings,
             dense_prompt_embeddings=dense_embeddings,
@@ -1497,13 +1502,13 @@ class SAM2VideoPredictor(SAM2Predictor):
             "pred_masks": torch.full(
                 size=(batch_size, 1, self.imgsz[0] // 4, self.imgsz[1] // 4),
                 fill_value=-1024.0,
-                dtype=torch.float32,
+                dtype=self.torch_dtype,
                 device=self.device,
             ),
             "obj_ptr": torch.full(
                 size=(batch_size, self.model.hidden_dim),
                 fill_value=-1024.0,
-                dtype=torch.float32,
+                dtype=self.torch_dtype,
                 device=self.device,
             ),
             "object_score_logits": torch.full(
@@ -1511,7 +1516,7 @@ class SAM2VideoPredictor(SAM2Predictor):
                 # default to 10.0 for object_score_logits, i.e. assuming the object is
                 # present as sigmoid(10)=1, same as in `predict_masks` of `MaskDecoder`
                 fill_value=10.0,
-                dtype=torch.float32,
+                dtype=self.torch_dtype,
                 device=self.device,
             ),
         }
@@ -1583,7 +1588,7 @@ class SAM2VideoPredictor(SAM2Predictor):
             feat_sizes=feat_sizes,
             point_inputs=None,
             # A dummy (empty) mask with a single object
-            mask_inputs=torch.zeros((1, 1, *self.imgsz), dtype=torch.float32, device=self.device),
+            mask_inputs=torch.zeros((1, 1, *self.imgsz), dtype=self.torch_dtype, device=self.device),
             output_dict={},
             num_frames=self.inference_state["num_frames"],
             track_in_reverse=False,

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/yolo/detect/val.py

@@ -204,11 +204,13 @@ class DetectionValidator(BaseValidator):
                 continue
 
             # Save
+            if self.args.save_json or self.args.save_txt:
+                predn_scaled = self.scale_preds(predn, pbatch)
             if self.args.save_json:
-                self.pred_to_json(predn, pbatch)
+                self.pred_to_json(predn_scaled, pbatch)
             if self.args.save_txt:
                 self.save_one_txt(
-                    predn,
+                    predn_scaled,
                     self.args.save_conf,
                     pbatch["ori_shape"],
                     self.save_dir / "labels" / f"{Path(pbatch['im_file']).stem}.txt",
@@ -373,13 +375,7 @@ class DetectionValidator(BaseValidator):
         """
         stem = Path(pbatch["im_file"]).stem
         image_id = int(stem) if stem.isnumeric() else stem
-        box = ops.scale_boxes(
-            pbatch["imgsz"],
-            predn["bboxes"].clone(),
-            pbatch["ori_shape"],
-            ratio_pad=pbatch["ratio_pad"],
-        )
-        box = ops.xyxy2xywh(box)  # xywh
+        box = ops.xyxy2xywh(predn["bboxes"])  # xywh
         box[:, :2] -= box[:, 2:] / 2  # xy center to top-left corner
         for b, s, c in zip(box.tolist(), predn["conf"].tolist(), predn["cls"].tolist()):
             self.jdict.append(
@@ -391,6 +387,18 @@ class DetectionValidator(BaseValidator):
                 }
             )
 
+    def scale_preds(self, predn: Dict[str, torch.Tensor], pbatch: Dict[str, Any]) -> Dict[str, torch.Tensor]:
+        """Scales predictions to the original image size."""
+        return {
+            **predn,
+            "bboxes": ops.scale_boxes(
+                pbatch["imgsz"],
+                predn["bboxes"].clone(),
+                pbatch["ori_shape"],
+                ratio_pad=pbatch["ratio_pad"],
+            ),
+        }
+
     def eval_json(self, stats: Dict[str, Any]) -> Dict[str, Any]:
         """
         Evaluate YOLO output in JSON format and return performance statistics.

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/yolo/model.py

@@ -370,7 +370,7 @@ class YOLOE(Model):
         stream: bool = False,
         visual_prompts: Dict[str, List] = {},
         refer_image=None,
-        predictor=None,
+        predictor=yolo.yoloe.YOLOEVPDetectPredictor,
         **kwargs,
     ):
         """
@@ -406,14 +406,16 @@ class YOLOE(Model):
                 f"Expected equal number of bounding boxes and classes, but got {len(visual_prompts['bboxes'])} and "
                 f"{len(visual_prompts['cls'])} respectively"
             )
-            if not isinstance(self.predictor, yolo.yoloe.YOLOEVPDetectPredictor):
-                self.predictor = (predictor or yolo.yoloe.YOLOEVPDetectPredictor)(
+            if type(self.predictor) is not predictor:
+                self.predictor = predictor(
                     overrides={
                         "task": self.model.task,
                         "mode": "predict",
                         "save": False,
                         "verbose": refer_image is None,
                         "batch": 1,
+                        "device": kwargs.get("device", None),
+                        "half": kwargs.get("half", False),
                     },
                     _callbacks=self.callbacks,
                 )

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/yolo/obb/val.py

@@ -179,9 +179,6 @@ class OBBValidator(DetectionValidator):
         stem = Path(pbatch["im_file"]).stem
         image_id = int(stem) if stem.isnumeric() else stem
         rbox = predn["bboxes"]
-        rbox = ops.scale_boxes(
-            pbatch["imgsz"], predn["bboxes"].clone(), pbatch["ori_shape"], ratio_pad=pbatch["ratio_pad"], xywh=True
-        )  # native-space pred
         poly = ops.xywhr2xyxyxyxy(rbox).view(-1, 8)
         for r, b, s, c in zip(rbox.tolist(), poly.tolist(), predn["conf"].tolist(), predn["cls"].tolist()):
             self.jdict.append(
@@ -221,6 +218,15 @@ class OBBValidator(DetectionValidator):
             obb=torch.cat([predn["bboxes"], predn["conf"].unsqueeze(-1), predn["cls"].unsqueeze(-1)], dim=1),
         ).save_txt(file, save_conf=save_conf)
 
+    def scale_preds(self, predn: Dict[str, torch.Tensor], pbatch: Dict[str, Any]) -> Dict[str, torch.Tensor]:
+        """Scales predictions to the original image size."""
+        return {
+            **predn,
+            "bboxes": ops.scale_boxes(
+                pbatch["imgsz"], predn["bboxes"].clone(), pbatch["ori_shape"], ratio_pad=pbatch["ratio_pad"], xywh=True
+            ),
+        }
+
     def eval_json(self, stats: Dict[str, Any]) -> Dict[str, Any]:
         """
         Evaluate YOLO output in JSON format and save predictions in DOTA format.

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/yolo/pose/val.py

@@ -242,15 +242,22 @@ class PoseValidator(DetectionValidator):
             before saving to the JSON dictionary.
         """
         super().pred_to_json(predn, pbatch)
-        kpts = ops.scale_coords(
-            pbatch["imgsz"],
-            predn["keypoints"].clone(),
-            pbatch["ori_shape"],
-            ratio_pad=pbatch["ratio_pad"],
-        )
+        kpts = predn["kpts"]
         for i, k in enumerate(kpts.flatten(1, 2).tolist()):
             self.jdict[-len(kpts) + i]["keypoints"] = k  # keypoints
 
+    def scale_preds(self, predn: Dict[str, torch.Tensor], pbatch: Dict[str, Any]) -> Dict[str, torch.Tensor]:
+        """Scales predictions to the original image size."""
+        return {
+            **super().scale_preds(predn, pbatch),
+            "kpts": ops.scale_coords(
+                pbatch["imgsz"],
+                predn["keypoints"].clone(),
+                pbatch["ori_shape"],
+                ratio_pad=pbatch["ratio_pad"],
+            ),
+        }
+
     def eval_json(self, stats: Dict[str, Any]) -> Dict[str, Any]:
         """Evaluate object detection model using COCO JSON format."""
         anno_json = self.data["path"] / "annotations/person_keypoints_val2017.json"  # annotations

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/yolo/segment/val.py

@@ -229,19 +229,24 @@ class SegmentationValidator(DetectionValidator):
             rle["counts"] = rle["counts"].decode("utf-8")
             return rle
 
-        coco_masks = torch.as_tensor(predn["masks"], dtype=torch.uint8)
-        coco_masks = ops.scale_image(
-            coco_masks.permute(1, 2, 0).contiguous().cpu().numpy(),
-            pbatch["ori_shape"],
-            ratio_pad=pbatch["ratio_pad"],
-        )
-        pred_masks = np.transpose(coco_masks, (2, 0, 1))
+        pred_masks = np.transpose(predn["masks"], (2, 0, 1))
         with ThreadPool(NUM_THREADS) as pool:
             rles = pool.map(single_encode, pred_masks)
         super().pred_to_json(predn, pbatch)
         for i, r in enumerate(rles):
             self.jdict[-len(rles) + i]["segmentation"] = r  # segmentation
 
+    def scale_preds(self, predn: Dict[str, torch.Tensor], pbatch: Dict[str, Any]) -> Dict[str, torch.Tensor]:
+        """Scales predictions to the original image size."""
+        return {
+            **super().scale_preds(predn, pbatch),
+            "masks": ops.scale_image(
+                torch.as_tensor(predn["masks"], dtype=torch.uint8).permute(1, 2, 0).contiguous().cpu().numpy(),
+                pbatch["ori_shape"],
+                ratio_pad=pbatch["ratio_pad"],
+            ),
+        }
+
     def eval_json(self, stats: Dict[str, Any]) -> Dict[str, Any]:
         """Return COCO-style instance segmentation evaluation metrics."""
         pred_json = self.save_dir / "predictions.json"  # predictions

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/models/yolo/yoloe/predict.py

@@ -71,7 +71,7 @@ class YOLOEVPDetectPredictor(DetectionPredictor):
         category = self.prompts["cls"]
         if len(img) == 1:
             visuals = self._process_single_image(img[0].shape[:2], im[0].shape[:2], category, bboxes, masks)
-            self.prompts = visuals.unsqueeze(0).to(self.device)  # (1, N, H, W)
+            prompts = visuals.unsqueeze(0).to(self.device)  # (1, N, H, W)
         else:
             # NOTE: only supports bboxes as prompts for now
             assert bboxes is not None, f"Expected bboxes, but got {bboxes}!"
@@ -89,8 +89,8 @@ class YOLOEVPDetectPredictor(DetectionPredictor):
                 self._process_single_image(img[i].shape[:2], im[i].shape[:2], category[i], bboxes[i])
                 for i in range(len(img))
             ]
-            self.prompts = torch.nn.utils.rnn.pad_sequence(visuals, batch_first=True).to(self.device)
-
+            prompts = torch.nn.utils.rnn.pad_sequence(visuals, batch_first=True).to(self.device)  # (B, N, H, W)
+        self.prompts = prompts.half() if self.model.fp16 else prompts.float()
         return img
 
     def _process_single_image(self, dst_shape, src_shape, category, bboxes=None, masks=None):

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/nn/modules/block.py

@@ -2025,9 +2025,7 @@ class SAVPE(nn.Module):
         vp = vp.reshape(B, Q, 1, -1)
 
         score = y * vp + torch.logical_not(vp) * torch.finfo(y.dtype).min
-
-        score = F.softmax(score, dim=-1, dtype=torch.float).to(score.dtype)
-
+        score = F.softmax(score, dim=-1).to(y.dtype)
         aggregated = score.transpose(-2, -3) @ x.reshape(B, self.c, C // self.c, -1).transpose(-1, -2)
 
         return F.normalize(aggregated.transpose(-2, -3).reshape(B, Q, -1), dim=-1, p=2)

{dgenerate_ultralytics_headless-8.3.180 → dgenerate_ultralytics_headless-8.3.181}/ultralytics/solutions/streamlit_inference.py

@@ -160,12 +160,19 @@ class Inference:
             ],
             key=lambda x: (M_ORD.index(x[:7].lower()), T_ORD.index(x[7:].lower() or "")),
         )
-        if self.model_path:  # If user provided the custom model, insert model without suffix as *.pt is added later
-            available_models.insert(0, self.model_path.split(".pt", 1)[0])
+        if self.model_path:  # Insert user provided custom model in available_models
+            available_models.insert(0, self.model_path)
         selected_model = self.st.sidebar.selectbox("Model", available_models)
 
         with self.st.spinner("Model is downloading..."):
-            self.model = YOLO(f"{selected_model.lower()}.pt")  # Load the YOLO model
+            if (
+                selected_model.endswith((".pt", ".onnx", ".torchscript", ".mlpackage", ".engine"))
+                or "openvino_model" in selected_model
+            ):
+                model_path = selected_model
+            else:
+                model_path = f"{selected_model.lower()}.pt"  # Default to .pt if no model provided during function call.
+            self.model = YOLO(model_path)  # Load the YOLO model
             class_names = list(self.model.names.values())  # Convert dictionary to list of class names
         self.st.success("Model loaded successfully!")