ultralytics 8.0.237__py3-none-any.whl → 8.0.239__py3-none-any.whl

ultralytics/models/sam/amg.py

@@ -8,10 +8,9 @@ import numpy as np
 import torch
 
 
-def is_box_near_crop_edge(boxes: torch.Tensor,
-                          crop_box: List[int],
-                          orig_box: List[int],
-                          atol: float = 20.0) -> torch.Tensor:
+def is_box_near_crop_edge(
+    boxes: torch.Tensor, crop_box: List[int], orig_box: List[int], atol: float = 20.0
+) -> torch.Tensor:
     """Return a boolean tensor indicating if boxes are near the crop edge."""
     crop_box_torch = torch.as_tensor(crop_box, dtype=torch.float, device=boxes.device)
     orig_box_torch = torch.as_tensor(orig_box, dtype=torch.float, device=boxes.device)
@@ -24,10 +23,10 @@ def is_box_near_crop_edge(boxes: torch.Tensor,
 
 def batch_iterator(batch_size: int, *args) -> Generator[List[Any], None, None]:
     """Yield batches of data from the input arguments."""
-    assert args and all(len(a) == len(args[0]) for a in args), 'Batched iteration must have same-size inputs.'
+    assert args and all(len(a) == len(args[0]) for a in args), "Batched iteration must have same-size inputs."
     n_batches = len(args[0]) // batch_size + int(len(args[0]) % batch_size != 0)
     for b in range(n_batches):
-        yield [arg[b * batch_size:(b + 1) * batch_size] for arg in args]
+        yield [arg[b * batch_size : (b + 1) * batch_size] for arg in args]
 
 
 def calculate_stability_score(masks: torch.Tensor, mask_threshold: float, threshold_offset: float) -> torch.Tensor:
@@ -39,9 +38,8 @@ def calculate_stability_score(masks: torch.Tensor, mask_threshold: float, thresh
     """
     # One mask is always contained inside the other.
     # Save memory by preventing unnecessary cast to torch.int64
-    intersections = ((masks > (mask_threshold + threshold_offset)).sum(-1, dtype=torch.int16).sum(-1,
-                                                                                                  dtype=torch.int32))
-    unions = ((masks > (mask_threshold - threshold_offset)).sum(-1, dtype=torch.int16).sum(-1, dtype=torch.int32))
+    intersections = (masks > (mask_threshold + threshold_offset)).sum(-1, dtype=torch.int16).sum(-1, dtype=torch.int32)
+    unions = (masks > (mask_threshold - threshold_offset)).sum(-1, dtype=torch.int16).sum(-1, dtype=torch.int32)
     return intersections / unions
 
 
@@ -56,11 +54,12 @@ def build_point_grid(n_per_side: int) -> np.ndarray:
 
 def build_all_layer_point_grids(n_per_side: int, n_layers: int, scale_per_layer: int) -> List[np.ndarray]:
     """Generate point grids for all crop layers."""
-    return [build_point_grid(int(n_per_side / (scale_per_layer ** i))) for i in range(n_layers + 1)]
+    return [build_point_grid(int(n_per_side / (scale_per_layer**i))) for i in range(n_layers + 1)]
 
 
-def generate_crop_boxes(im_size: Tuple[int, ...], n_layers: int,
-                        overlap_ratio: float) -> Tuple[List[List[int]], List[int]]:
+def generate_crop_boxes(
+    im_size: Tuple[int, ...], n_layers: int, overlap_ratio: float
+) -> Tuple[List[List[int]], List[int]]:
     """
     Generates a list of crop boxes of different sizes.
 
@@ -132,8 +131,8 @@ def remove_small_regions(mask: np.ndarray, area_thresh: float, mode: str) -> Tup
     """Remove small disconnected regions or holes in a mask, returning the mask and a modification indicator."""
     import cv2  # type: ignore
 
-    assert mode in {'holes', 'islands'}
-    correct_holes = mode == 'holes'
+    assert mode in {"holes", "islands"}
+    correct_holes = mode == "holes"
     working_mask = (correct_holes ^ mask).astype(np.uint8)
     n_labels, regions, stats, _ = cv2.connectedComponentsWithStats(working_mask, 8)
     sizes = stats[:, -1][1:]  # Row 0 is background label

ultralytics/models/sam/build.py

@@ -64,46 +64,47 @@ def build_mobile_sam(checkpoint=None):
     )
 
 
-def _build_sam(encoder_embed_dim,
-               encoder_depth,
-               encoder_num_heads,
-               encoder_global_attn_indexes,
-               checkpoint=None,
-               mobile_sam=False):
+def _build_sam(
+    encoder_embed_dim, encoder_depth, encoder_num_heads, encoder_global_attn_indexes, checkpoint=None, mobile_sam=False
+):
     """Builds the selected SAM model architecture."""
     prompt_embed_dim = 256
     image_size = 1024
     vit_patch_size = 16
     image_embedding_size = image_size // vit_patch_size
-    image_encoder = (TinyViT(
-        img_size=1024,
-        in_chans=3,
-        num_classes=1000,
-        embed_dims=encoder_embed_dim,
-        depths=encoder_depth,
-        num_heads=encoder_num_heads,
-        window_sizes=[7, 7, 14, 7],
-        mlp_ratio=4.0,
-        drop_rate=0.0,
-        drop_path_rate=0.0,
-        use_checkpoint=False,
-        mbconv_expand_ratio=4.0,
-        local_conv_size=3,
-        layer_lr_decay=0.8,
-    ) if mobile_sam else ImageEncoderViT(
-        depth=encoder_depth,
-        embed_dim=encoder_embed_dim,
-        img_size=image_size,
-        mlp_ratio=4,
-        norm_layer=partial(torch.nn.LayerNorm, eps=1e-6),
-        num_heads=encoder_num_heads,
-        patch_size=vit_patch_size,
-        qkv_bias=True,
-        use_rel_pos=True,
-        global_attn_indexes=encoder_global_attn_indexes,
-        window_size=14,
-        out_chans=prompt_embed_dim,
-    ))
+    image_encoder = (
+        TinyViT(
+            img_size=1024,
+            in_chans=3,
+            num_classes=1000,
+            embed_dims=encoder_embed_dim,
+            depths=encoder_depth,
+            num_heads=encoder_num_heads,
+            window_sizes=[7, 7, 14, 7],
+            mlp_ratio=4.0,
+            drop_rate=0.0,
+            drop_path_rate=0.0,
+            use_checkpoint=False,
+            mbconv_expand_ratio=4.0,
+            local_conv_size=3,
+            layer_lr_decay=0.8,
+        )
+        if mobile_sam
+        else ImageEncoderViT(
+            depth=encoder_depth,
+            embed_dim=encoder_embed_dim,
+            img_size=image_size,
+            mlp_ratio=4,
+            norm_layer=partial(torch.nn.LayerNorm, eps=1e-6),
+            num_heads=encoder_num_heads,
+            patch_size=vit_patch_size,
+            qkv_bias=True,
+            use_rel_pos=True,
+            global_attn_indexes=encoder_global_attn_indexes,
+            window_size=14,
+            out_chans=prompt_embed_dim,
+        )
+    )
     sam = Sam(
         image_encoder=image_encoder,
         prompt_encoder=PromptEncoder(
@@ -129,7 +130,7 @@ def _build_sam(encoder_embed_dim,
     )
     if checkpoint is not None:
         checkpoint = attempt_download_asset(checkpoint)
-        with open(checkpoint, 'rb') as f:
+        with open(checkpoint, "rb") as f:
             state_dict = torch.load(f)
         sam.load_state_dict(state_dict)
     sam.eval()
@@ -139,13 +140,14 @@ def _build_sam(encoder_embed_dim,
 
 
 sam_model_map = {
-    'sam_h.pt': build_sam_vit_h,
-    'sam_l.pt': build_sam_vit_l,
-    'sam_b.pt': build_sam_vit_b,
-    'mobile_sam.pt': build_mobile_sam, }
+    "sam_h.pt": build_sam_vit_h,
+    "sam_l.pt": build_sam_vit_l,
+    "sam_b.pt": build_sam_vit_b,
+    "mobile_sam.pt": build_mobile_sam,
+}
 
 
-def build_sam(ckpt='sam_b.pt'):
+def build_sam(ckpt="sam_b.pt"):
     """Build a SAM model specified by ckpt."""
     model_builder = None
     ckpt = str(ckpt)  # to allow Path ckpt types
@@ -154,6 +156,6 @@ def build_sam(ckpt='sam_b.pt'):
             model_builder = sam_model_map.get(k)
 
     if not model_builder:
-        raise FileNotFoundError(f'{ckpt} is not a supported SAM model. Available models are: \n {sam_model_map.keys()}')
+        raise FileNotFoundError(f"{ckpt} is not a supported SAM model. Available models are: \n {sam_model_map.keys()}")
 
     return model_builder(ckpt)

ultralytics/models/sam/model.py

@@ -32,7 +32,7 @@ class SAM(Model):
     dataset.
     """
 
-    def __init__(self, model='sam_b.pt') -> None:
+    def __init__(self, model="sam_b.pt") -> None:
         """
         Initializes the SAM model with a pre-trained model file.
 
@@ -42,9 +42,9 @@ class SAM(Model):
         Raises:
             NotImplementedError: If the model file extension is not .pt or .pth.
         """
-        if model and Path(model).suffix not in ('.pt', '.pth'):
-            raise NotImplementedError('SAM prediction requires pre-trained *.pt or *.pth model.')
-        super().__init__(model=model, task='segment')
+        if model and Path(model).suffix not in (".pt", ".pth"):
+            raise NotImplementedError("SAM prediction requires pre-trained *.pt or *.pth model.")
+        super().__init__(model=model, task="segment")
 
     def _load(self, weights: str, task=None):
         """
@@ -70,7 +70,7 @@ class SAM(Model):
         Returns:
             (list): The model predictions.
         """
-        overrides = dict(conf=0.25, task='segment', mode='predict', imgsz=1024)
+        overrides = dict(conf=0.25, task="segment", mode="predict", imgsz=1024)
         kwargs.update(overrides)
         prompts = dict(bboxes=bboxes, points=points, labels=labels)
         return super().predict(source, stream, prompts=prompts, **kwargs)
@@ -112,4 +112,4 @@ class SAM(Model):
         Returns:
             (dict): A dictionary mapping the 'segment' task to its corresponding 'Predictor'.
         """
-        return {'segment': {'predictor': Predictor}}
+        return {"segment": {"predictor": Predictor}}

ultralytics/models/sam/modules/decoders.py

@@ -64,8 +64,9 @@ class MaskDecoder(nn.Module):
             nn.ConvTranspose2d(transformer_dim // 4, transformer_dim // 8, kernel_size=2, stride=2),
             activation(),
         )
-        self.output_hypernetworks_mlps = nn.ModuleList([
-            MLP(transformer_dim, transformer_dim, transformer_dim // 8, 3) for _ in range(self.num_mask_tokens)])
+        self.output_hypernetworks_mlps = nn.ModuleList(
+            [MLP(transformer_dim, transformer_dim, transformer_dim // 8, 3) for _ in range(self.num_mask_tokens)]
+        )
 
         self.iou_prediction_head = MLP(transformer_dim, iou_head_hidden_dim, self.num_mask_tokens, iou_head_depth)
 
@@ -132,13 +133,14 @@ class MaskDecoder(nn.Module):
         # Run the transformer
         hs, src = self.transformer(src, pos_src, tokens)
         iou_token_out = hs[:, 0, :]
-        mask_tokens_out = hs[:, 1:(1 + self.num_mask_tokens), :]
+        mask_tokens_out = hs[:, 1 : (1 + self.num_mask_tokens), :]
 
         # Upscale mask embeddings and predict masks using the mask tokens
         src = src.transpose(1, 2).view(b, c, h, w)
         upscaled_embedding = self.output_upscaling(src)
         hyper_in_list: List[torch.Tensor] = [
-            self.output_hypernetworks_mlps[i](mask_tokens_out[:, i, :]) for i in range(self.num_mask_tokens)]
+            self.output_hypernetworks_mlps[i](mask_tokens_out[:, i, :]) for i in range(self.num_mask_tokens)
+        ]
         hyper_in = torch.stack(hyper_in_list, dim=1)
         b, c, h, w = upscaled_embedding.shape
         masks = (hyper_in @ upscaled_embedding.view(b, c, h * w)).view(b, -1, h, w)

ultralytics/models/sam/modules/encoders.py

@@ -28,23 +28,23 @@ class ImageEncoderViT(nn.Module):
     """
 
     def __init__(
-            self,
-            img_size: int = 1024,
-            patch_size: int = 16,
-            in_chans: int = 3,
-            embed_dim: int = 768,
-            depth: int = 12,
-            num_heads: int = 12,
-            mlp_ratio: float = 4.0,
-            out_chans: int = 256,
-            qkv_bias: bool = True,
-            norm_layer: Type[nn.Module] = nn.LayerNorm,
-            act_layer: Type[nn.Module] = nn.GELU,
-            use_abs_pos: bool = True,
-            use_rel_pos: bool = False,
-            rel_pos_zero_init: bool = True,
-            window_size: int = 0,
-            global_attn_indexes: Tuple[int, ...] = (),
+        self,
+        img_size: int = 1024,
+        patch_size: int = 16,
+        in_chans: int = 3,
+        embed_dim: int = 768,
+        depth: int = 12,
+        num_heads: int = 12,
+        mlp_ratio: float = 4.0,
+        out_chans: int = 256,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = nn.GELU,
+        use_abs_pos: bool = True,
+        use_rel_pos: bool = False,
+        rel_pos_zero_init: bool = True,
+        window_size: int = 0,
+        global_attn_indexes: Tuple[int, ...] = (),
     ) -> None:
         """
         Args:
@@ -283,9 +283,9 @@ class PromptEncoder(nn.Module):
         if masks is not None:
             dense_embeddings = self._embed_masks(masks)
         else:
-            dense_embeddings = self.no_mask_embed.weight.reshape(1, -1, 1,
-                                                                 1).expand(bs, -1, self.image_embedding_size[0],
-                                                                           self.image_embedding_size[1])
+            dense_embeddings = self.no_mask_embed.weight.reshape(1, -1, 1, 1).expand(
+                bs, -1, self.image_embedding_size[0], self.image_embedding_size[1]
+            )
 
         return sparse_embeddings, dense_embeddings
 
@@ -298,7 +298,7 @@ class PositionEmbeddingRandom(nn.Module):
         super().__init__()
         if scale is None or scale <= 0.0:
             scale = 1.0
-        self.register_buffer('positional_encoding_gaussian_matrix', scale * torch.randn((2, num_pos_feats)))
+        self.register_buffer("positional_encoding_gaussian_matrix", scale * torch.randn((2, num_pos_feats)))
 
         # Set non-deterministic for forward() error 'cumsum_cuda_kernel does not have a deterministic implementation'
         torch.use_deterministic_algorithms(False)
@@ -425,14 +425,14 @@ class Attention(nn.Module):
         super().__init__()
         self.num_heads = num_heads
         head_dim = dim // num_heads
-        self.scale = head_dim ** -0.5
+        self.scale = head_dim**-0.5
 
         self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
         self.proj = nn.Linear(dim, dim)
 
         self.use_rel_pos = use_rel_pos
         if self.use_rel_pos:
-            assert (input_size is not None), 'Input size must be provided if using relative positional encoding.'
+            assert input_size is not None, "Input size must be provided if using relative positional encoding."
             # Initialize relative positional embeddings
             self.rel_pos_h = nn.Parameter(torch.zeros(2 * input_size[0] - 1, head_dim))
             self.rel_pos_w = nn.Parameter(torch.zeros(2 * input_size[1] - 1, head_dim))
@@ -479,8 +479,9 @@ def window_partition(x: torch.Tensor, window_size: int) -> Tuple[torch.Tensor, T
     return windows, (Hp, Wp)
 
 
-def window_unpartition(windows: torch.Tensor, window_size: int, pad_hw: Tuple[int, int],
-                       hw: Tuple[int, int]) -> torch.Tensor:
+def window_unpartition(
+    windows: torch.Tensor, window_size: int, pad_hw: Tuple[int, int], hw: Tuple[int, int]
+) -> torch.Tensor:
     """
     Window unpartition into original sequences and removing padding.
 
@@ -523,7 +524,7 @@ def get_rel_pos(q_size: int, k_size: int, rel_pos: torch.Tensor) -> torch.Tensor
         rel_pos_resized = F.interpolate(
             rel_pos.reshape(1, rel_pos.shape[0], -1).permute(0, 2, 1),
             size=max_rel_dist,
-            mode='linear',
+            mode="linear",
         )
         rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).permute(1, 0)
     else:
@@ -567,11 +568,12 @@ def add_decomposed_rel_pos(
 
     B, _, dim = q.shape
     r_q = q.reshape(B, q_h, q_w, dim)
-    rel_h = torch.einsum('bhwc,hkc->bhwk', r_q, Rh)
-    rel_w = torch.einsum('bhwc,wkc->bhwk', r_q, Rw)
+    rel_h = torch.einsum("bhwc,hkc->bhwk", r_q, Rh)
+    rel_w = torch.einsum("bhwc,wkc->bhwk", r_q, Rw)
 
     attn = (attn.view(B, q_h, q_w, k_h, k_w) + rel_h[:, :, :, :, None] + rel_w[:, :, :, None, :]).view(
-        B, q_h * q_w, k_h * k_w)
+        B, q_h * q_w, k_h * k_w
+    )
 
     return attn
 
@@ -580,12 +582,12 @@ class PatchEmbed(nn.Module):
     """Image to Patch Embedding."""
 
     def __init__(
-            self,
-            kernel_size: Tuple[int, int] = (16, 16),
-            stride: Tuple[int, int] = (16, 16),
-            padding: Tuple[int, int] = (0, 0),
-            in_chans: int = 3,
-            embed_dim: int = 768,
+        self,
+        kernel_size: Tuple[int, int] = (16, 16),
+        stride: Tuple[int, int] = (16, 16),
+        padding: Tuple[int, int] = (0, 0),
+        in_chans: int = 3,
+        embed_dim: int = 768,
     ) -> None:
         """
         Initialize PatchEmbed module.

ultralytics/models/sam/modules/sam.py

@@ -30,8 +30,9 @@ class Sam(nn.Module):
         pixel_mean (List[float]): Mean pixel values for image normalization.
         pixel_std (List[float]): Standard deviation values for image normalization.
     """
+
     mask_threshold: float = 0.0
-    image_format: str = 'RGB'
+    image_format: str = "RGB"
 
     def __init__(
         self,
@@ -39,7 +40,7 @@ class Sam(nn.Module):
         prompt_encoder: PromptEncoder,
         mask_decoder: MaskDecoder,
         pixel_mean: List[float] = (123.675, 116.28, 103.53),
-        pixel_std: List[float] = (58.395, 57.12, 57.375)
+        pixel_std: List[float] = (58.395, 57.12, 57.375),
     ) -> None:
         """
         Initialize the Sam class to predict object masks from an image and input prompts.
@@ -60,5 +61,5 @@ class Sam(nn.Module):
         self.image_encoder = image_encoder
         self.prompt_encoder = prompt_encoder
         self.mask_decoder = mask_decoder
-        self.register_buffer('pixel_mean', torch.Tensor(pixel_mean).view(-1, 1, 1), False)
-        self.register_buffer('pixel_std', torch.Tensor(pixel_std).view(-1, 1, 1), False)
+        self.register_buffer("pixel_mean", torch.Tensor(pixel_mean).view(-1, 1, 1), False)
+        self.register_buffer("pixel_std", torch.Tensor(pixel_std).view(-1, 1, 1), False)