Sync branch 'mashb1t_main' with develop_upstream

extras/GroundingDINO/config/GroundingDINO_SwinT_OGC.py

@@ -0,0 +1,43 @@
+batch_size = 1
+modelname = "groundingdino"
+backbone = "swin_T_224_1k"
+position_embedding = "sine"
+pe_temperatureH = 20
+pe_temperatureW = 20
+return_interm_indices = [1, 2, 3]
+backbone_freeze_keywords = None
+enc_layers = 6
+dec_layers = 6
+pre_norm = False
+dim_feedforward = 2048
+hidden_dim = 256
+dropout = 0.0
+nheads = 8
+num_queries = 900
+query_dim = 4
+num_patterns = 0
+num_feature_levels = 4
+enc_n_points = 4
+dec_n_points = 4
+two_stage_type = "standard"
+two_stage_bbox_embed_share = False
+two_stage_class_embed_share = False
+transformer_activation = "relu"
+dec_pred_bbox_embed_share = True
+dn_box_noise_scale = 1.0
+dn_label_noise_ratio = 0.5
+dn_label_coef = 1.0
+dn_bbox_coef = 1.0
+embed_init_tgt = True
+dn_labelbook_size = 2000
+max_text_len = 256
+text_encoder_type = "bert-base-uncased"
+use_text_enhancer = True
+use_fusion_layer = True
+use_checkpoint = True
+use_transformer_ckpt = True
+use_text_cross_attention = True
+text_dropout = 0.0
+fusion_dropout = 0.0
+fusion_droppath = 0.1
+sub_sentence_present = True

extras/GroundingDINO/util/inference.py

@@ -0,0 +1,100 @@
+from typing import Tuple, List
+
+import ldm_patched.modules.model_management as model_management
+from ldm_patched.modules.model_patcher import ModelPatcher
+from modules.config import path_inpaint
+from modules.model_loader import load_file_from_url
+
+import numpy as np
+import supervision as sv
+import torch
+from groundingdino.util.inference import Model
+from groundingdino.util.inference import load_model, preprocess_caption, get_phrases_from_posmap
+
+
+class GroundingDinoModel(Model):
+    def __init__(self):
+        self.config_file = 'extras/GroundingDINO/config/GroundingDINO_SwinT_OGC.py'
+        self.model = None
+        self.load_device = torch.device('cpu')
+        self.offload_device = torch.device('cpu')
+
+    @torch.no_grad()
+    @torch.inference_mode()
+    def predict_with_caption(
+            self,
+            image: np.ndarray,
+            caption: str,
+            box_threshold: float = 0.35,
+            text_threshold: float = 0.25
+    ) -> Tuple[sv.Detections, torch.Tensor, torch.Tensor, List[str]]:
+        if self.model is None:
+            filename = load_file_from_url(
+                url="https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth",
+                file_name='groundingdino_swint_ogc.pth',
+                model_dir=path_inpaint)
+            model = load_model(model_config_path=self.config_file, model_checkpoint_path=filename)
+
+            self.load_device = model_management.text_encoder_device()
+            self.offload_device = model_management.text_encoder_offload_device()
+
+            model.to(self.offload_device)
+
+            self.model = ModelPatcher(model, load_device=self.load_device, offload_device=self.offload_device)
+
+        model_management.load_model_gpu(self.model)
+
+        processed_image = GroundingDinoModel.preprocess_image(image_bgr=image).to(self.load_device)
+        boxes, logits, phrases = predict(
+            model=self.model,
+            image=processed_image,
+            caption=caption,
+            box_threshold=box_threshold,
+            text_threshold=text_threshold,
+            device=self.load_device)
+        source_h, source_w, _ = image.shape
+        detections = GroundingDinoModel.post_process_result(
+            source_h=source_h,
+            source_w=source_w,
+            boxes=boxes,
+            logits=logits)
+        return detections, boxes, logits, phrases
+
+
+def predict(
+        model,
+        image: torch.Tensor,
+        caption: str,
+        box_threshold: float,
+        text_threshold: float,
+        device: str = "cuda"
+) -> Tuple[torch.Tensor, torch.Tensor, List[str]]:
+    caption = preprocess_caption(caption=caption)
+
+    # override to use model wrapped by patcher
+    model = model.model.to(device)
+    image = image.to(device)
+
+    with torch.no_grad():
+        outputs = model(image[None], captions=[caption])
+
+    prediction_logits = outputs["pred_logits"].cpu().sigmoid()[0]  # prediction_logits.shape = (nq, 256)
+    prediction_boxes = outputs["pred_boxes"].cpu()[0]  # prediction_boxes.shape = (nq, 4)
+
+    mask = prediction_logits.max(dim=1)[0] > box_threshold
+    logits = prediction_logits[mask]  # logits.shape = (n, 256)
+    boxes = prediction_boxes[mask]  # boxes.shape = (n, 4)
+
+    tokenizer = model.tokenizer
+    tokenized = tokenizer(caption)
+
+    phrases = [
+        get_phrases_from_posmap(logit > text_threshold, tokenized, tokenizer).replace('.', '')
+        for logit
+        in logits
+    ]
+
+    return boxes, logits.max(dim=1)[0], phrases
+
+
+default_groundingdino = GroundingDinoModel().predict_with_caption

extras/censor.py

@@ -41,7 +41,7 @@ class Censor:
         model_management.load_model_gpu(self.safety_checker_model)
 
         single = False
-        if not isinstance(images, list) or isinstance(images, np.ndarray):
+        if not isinstance(images, (list, np.ndarray)):
             images = [images]
             single = True
 

extras/inpaint_mask.py

@@ -0,0 +1,130 @@
+import sys
+
+import modules.config
+import numpy as np
+import torch
+from extras.GroundingDINO.util.inference import default_groundingdino
+from extras.sam.predictor import SamPredictor
+from rembg import remove, new_session
+from segment_anything import sam_model_registry
+from segment_anything.utils.amg import remove_small_regions
+
+
+class SAMOptions:
+    def __init__(self,
+                 # GroundingDINO
+                 dino_prompt: str = '',
+                 dino_box_threshold=0.3,
+                 dino_text_threshold=0.25,
+                 dino_erode_or_dilate=0,
+                 dino_debug=False,
+
+                 # SAM
+                 max_detections=2,
+                 model_type='vit_b'
+                 ):
+        self.dino_prompt = dino_prompt
+        self.dino_box_threshold = dino_box_threshold
+        self.dino_text_threshold = dino_text_threshold
+        self.dino_erode_or_dilate = dino_erode_or_dilate
+        self.dino_debug = dino_debug
+        self.max_detections = max_detections
+        self.model_type = model_type
+
+
+def optimize_masks(masks: torch.Tensor) -> torch.Tensor:
+    """
+    removes small disconnected regions and holes
+    """
+    fine_masks = []
+    for mask in masks.to('cpu').numpy():  # masks: [num_masks, 1, h, w]
+        fine_masks.append(remove_small_regions(mask[0], 400, mode="holes")[0])
+    masks = np.stack(fine_masks, axis=0)[:, np.newaxis]
+    return torch.from_numpy(masks)
+
+
+def generate_mask_from_image(image: np.ndarray, mask_model: str = 'sam', extras=None,
+                             sam_options: SAMOptions | None = SAMOptions) -> tuple[np.ndarray | None, int | None, int | None, int | None]:
+    dino_detection_count = 0
+    sam_detection_count = 0
+    sam_detection_on_mask_count = 0
+
+    if image is None:
+        return None, dino_detection_count, sam_detection_count, sam_detection_on_mask_count
+
+    if extras is None:
+        extras = {}
+
+    if 'image' in image:
+        image = image['image']
+
+    if mask_model != 'sam' or sam_options is None:
+        result = remove(
+            image,
+            session=new_session(mask_model, **extras),
+            only_mask=True,
+            **extras
+        )
+
+        return result, dino_detection_count, sam_detection_count, sam_detection_on_mask_count
+
+    detections, boxes, logits, phrases = default_groundingdino(
+        image=image,
+        caption=sam_options.dino_prompt,
+        box_threshold=sam_options.dino_box_threshold,
+        text_threshold=sam_options.dino_text_threshold
+    )
+
+    H, W = image.shape[0], image.shape[1]
+    boxes = boxes * torch.Tensor([W, H, W, H])
+    boxes[:, :2] = boxes[:, :2] - boxes[:, 2:] / 2
+    boxes[:, 2:] = boxes[:, 2:] + boxes[:, :2]
+
+    sam_checkpoint = modules.config.download_sam_model(sam_options.model_type)
+    sam = sam_model_registry[sam_options.model_type](checkpoint=sam_checkpoint)
+
+    sam_predictor = SamPredictor(sam)
+    final_mask_tensor = torch.zeros((image.shape[0], image.shape[1]))
+    dino_detection_count = boxes.size(0)
+
+    if dino_detection_count > 0:
+        sam_predictor.set_image(image)
+
+        if sam_options.dino_erode_or_dilate != 0:
+            for index in range(boxes.size(0)):
+                assert boxes.size(1) == 4
+                boxes[index][0] -= sam_options.dino_erode_or_dilate
+                boxes[index][1] -= sam_options.dino_erode_or_dilate
+                boxes[index][2] += sam_options.dino_erode_or_dilate
+                boxes[index][3] += sam_options.dino_erode_or_dilate
+
+        if sam_options.dino_debug:
+            from PIL import ImageDraw, Image
+            debug_dino_image = Image.new("RGB", (image.shape[1], image.shape[0]), color="black")
+            draw = ImageDraw.Draw(debug_dino_image)
+            for box in boxes.numpy():
+                draw.rectangle(box.tolist(), fill="white")
+            return np.array(debug_dino_image), dino_detection_count, sam_detection_count, sam_detection_on_mask_count
+
+        transformed_boxes = sam_predictor.transform.apply_boxes_torch(boxes, image.shape[:2])
+        masks, _, _ = sam_predictor.predict_torch(
+            point_coords=None,
+            point_labels=None,
+            boxes=transformed_boxes,
+            multimask_output=False,
+        )
+
+        masks = optimize_masks(masks)
+        sam_detection_count = len(masks)
+        if sam_options.max_detections == 0:
+            sam_options.max_detections = sys.maxsize
+        sam_objects = min(len(logits), sam_options.max_detections)
+        for obj_ind in range(sam_objects):
+            mask_tensor = masks[obj_ind][0]
+            final_mask_tensor += mask_tensor
+            sam_detection_on_mask_count += 1
+
+    final_mask_tensor = (final_mask_tensor > 0).to('cpu').numpy()
+    mask_image = np.dstack((final_mask_tensor, final_mask_tensor, final_mask_tensor)) * 255
+    mask_image = np.array(mask_image, dtype=np.uint8)
+    return mask_image, dino_detection_count, sam_detection_count, sam_detection_on_mask_count

extras/sam/predictor.py

@@ -0,0 +1,288 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+
+# This source code is licensed under the license found in the
+# LICENSE file in the root directory of this source tree.
+
+import numpy as np
+import torch
+from ldm_patched.modules import model_management
+from ldm_patched.modules.model_patcher import ModelPatcher
+
+from segment_anything.modeling import Sam
+
+from typing import Optional, Tuple
+
+from segment_anything.utils.transforms import ResizeLongestSide
+
+
+class SamPredictor:
+    def __init__(
+        self,
+        model: Sam,
+        load_device=model_management.text_encoder_device(),
+        offload_device=model_management.text_encoder_offload_device()
+    ) -> None:
+        """
+        Uses SAM to calculate the image embedding for an image, and then
+        allow repeated, efficient mask prediction given prompts.
+
+        Arguments:
+          model (Sam): The model to use for mask prediction.
+        """
+        super().__init__()
+
+        self.load_device = load_device
+        self.offload_device = offload_device
+        # can't use model.half() here as slow_conv2d_cpu is not implemented for half
+        model.to(self.offload_device)
+
+        self.patcher = ModelPatcher(model, load_device=self.load_device, offload_device=self.offload_device)
+
+        self.transform = ResizeLongestSide(model.image_encoder.img_size)
+        self.reset_image()
+
+    def set_image(
+        self,
+        image: np.ndarray,
+        image_format: str = "RGB",
+    ) -> None:
+        """
+        Calculates the image embeddings for the provided image, allowing
+        masks to be predicted with the 'predict' method.
+
+        Arguments:
+          image (np.ndarray): The image for calculating masks. Expects an
+            image in HWC uint8 format, with pixel values in [0, 255].
+          image_format (str): The color format of the image, in ['RGB', 'BGR'].
+        """
+        assert image_format in [
+            "RGB",
+            "BGR",
+        ], f"image_format must be in ['RGB', 'BGR'], is {image_format}."
+        if image_format != self.patcher.model.image_format:
+            image = image[..., ::-1]
+
+        # Transform the image to the form expected by the model
+        input_image = self.transform.apply_image(image)
+        input_image_torch = torch.as_tensor(input_image, device=self.load_device)
+        input_image_torch = input_image_torch.permute(2, 0, 1).contiguous()[None, :, :, :]
+
+        self.set_torch_image(input_image_torch, image.shape[:2])
+
+    @torch.no_grad()
+    def set_torch_image(
+        self,
+        transformed_image: torch.Tensor,
+        original_image_size: Tuple[int, ...],
+    ) -> None:
+        """
+        Calculates the image embeddings for the provided image, allowing
+        masks to be predicted with the 'predict' method. Expects the input
+        image to be already transformed to the format expected by the model.
+
+        Arguments:
+          transformed_image (torch.Tensor): The input image, with shape
+            1x3xHxW, which has been transformed with ResizeLongestSide.
+          original_image_size (tuple(int, int)): The size of the image
+            before transformation, in (H, W) format.
+        """
+        assert (
+            len(transformed_image.shape) == 4
+            and transformed_image.shape[1] == 3
+            and max(*transformed_image.shape[2:]) == self.patcher.model.image_encoder.img_size
+        ), f"set_torch_image input must be BCHW with long side {self.patcher.model.image_encoder.img_size}."
+        self.reset_image()
+
+        self.original_size = original_image_size
+        self.input_size = tuple(transformed_image.shape[-2:])
+        model_management.load_model_gpu(self.patcher)
+        input_image = self.patcher.model.preprocess(transformed_image.to(self.load_device))
+        self.features = self.patcher.model.image_encoder(input_image)
+        self.is_image_set = True
+
+    def predict(
+        self,
+        point_coords: Optional[np.ndarray] = None,
+        point_labels: Optional[np.ndarray] = None,
+        box: Optional[np.ndarray] = None,
+        mask_input: Optional[np.ndarray] = None,
+        multimask_output: bool = True,
+        return_logits: bool = False,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        Predict masks for the given input prompts, using the currently set image.
+
+        Arguments:
+          point_coords (np.ndarray or None): A Nx2 array of point prompts to the
+            model. Each point is in (X,Y) in pixels.
+          point_labels (np.ndarray or None): A length N array of labels for the
+            point prompts. 1 indicates a foreground point and 0 indicates a
+            background point.
+          box (np.ndarray or None): A length 4 array given a box prompt to the
+            model, in XYXY format.
+          mask_input (np.ndarray): A low resolution mask input to the model, typically
+            coming from a previous prediction iteration. Has form 1xHxW, where
+            for SAM, H=W=256.
+          multimask_output (bool): If true, the model will return three masks.
+            For ambiguous input prompts (such as a single click), this will often
+            produce better masks than a single prediction. If only a single
+            mask is needed, the model's predicted quality score can be used
+            to select the best mask. For non-ambiguous prompts, such as multiple
+            input prompts, multimask_output=False can give better results.
+          return_logits (bool): If true, returns un-thresholded masks logits
+            instead of a binary mask.
+
+        Returns:
+          (np.ndarray): The output masks in CxHxW format, where C is the
+            number of masks, and (H, W) is the original image size.
+          (np.ndarray): An array of length C containing the model's
+            predictions for the quality of each mask.
+          (np.ndarray): An array of shape CxHxW, where C is the number
+            of masks and H=W=256. These low resolution logits can be passed to
+            a subsequent iteration as mask input.
+        """
+        if not self.is_image_set:
+            raise RuntimeError("An image must be set with .set_image(...) before mask prediction.")
+
+        # Transform input prompts
+        coords_torch, labels_torch, box_torch, mask_input_torch = None, None, None, None
+        if point_coords is not None:
+            assert (
+                point_labels is not None
+            ), "point_labels must be supplied if point_coords is supplied."
+            point_coords = self.transform.apply_coords(point_coords, self.original_size)
+            coords_torch = torch.as_tensor(point_coords, dtype=torch.float, device=self.load_device)
+            labels_torch = torch.as_tensor(point_labels, dtype=torch.int, device=self.load_device)
+            coords_torch, labels_torch = coords_torch[None, :, :], labels_torch[None, :]
+        if box is not None:
+            box = self.transform.apply_boxes(box, self.original_size)
+            box_torch = torch.as_tensor(box, dtype=torch.float, device=self.load_device)
+            box_torch = box_torch[None, :]
+        if mask_input is not None:
+            mask_input_torch = torch.as_tensor(mask_input, dtype=torch.float, device=self.load_device)
+            mask_input_torch = mask_input_torch[None, :, :, :]
+
+        masks, iou_predictions, low_res_masks = self.predict_torch(
+            coords_torch,
+            labels_torch,
+            box_torch,
+            mask_input_torch,
+            multimask_output,
+            return_logits=return_logits,
+        )
+
+        masks = masks[0].detach().cpu().numpy()
+        iou_predictions = iou_predictions[0].detach().cpu().numpy()
+        low_res_masks = low_res_masks[0].detach().cpu().numpy()
+        return masks, iou_predictions, low_res_masks
+
+    @torch.no_grad()
+    def predict_torch(
+        self,
+        point_coords: Optional[torch.Tensor],
+        point_labels: Optional[torch.Tensor],
+        boxes: Optional[torch.Tensor] = None,
+        mask_input: Optional[torch.Tensor] = None,
+        multimask_output: bool = True,
+        return_logits: bool = False,
+    ) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+        """
+        Predict masks for the given input prompts, using the currently set image.
+        Input prompts are batched torch tensors and are expected to already be
+        transformed to the input frame using ResizeLongestSide.
+
+        Arguments:
+          point_coords (torch.Tensor or None): A BxNx2 array of point prompts to the
+            model. Each point is in (X,Y) in pixels.
+          point_labels (torch.Tensor or None): A BxN array of labels for the
+            point prompts. 1 indicates a foreground point and 0 indicates a
+            background point.
+          box (np.ndarray or None): A Bx4 array given a box prompt to the
+            model, in XYXY format.
+          mask_input (np.ndarray): A low resolution mask input to the model, typically
+            coming from a previous prediction iteration. Has form Bx1xHxW, where
+            for SAM, H=W=256. Masks returned by a previous iteration of the
+            predict method do not need further transformation.
+          multimask_output (bool): If true, the model will return three masks.
+            For ambiguous input prompts (such as a single click), this will often
+            produce better masks than a single prediction. If only a single
+            mask is needed, the model's predicted quality score can be used
+            to select the best mask. For non-ambiguous prompts, such as multiple
+            input prompts, multimask_output=False can give better results.
+          return_logits (bool): If true, returns un-thresholded masks logits
+            instead of a binary mask.
+
+        Returns:
+          (torch.Tensor): The output masks in BxCxHxW format, where C is the
+            number of masks, and (H, W) is the original image size.
+          (torch.Tensor): An array of shape BxC containing the model's
+            predictions for the quality of each mask.
+          (torch.Tensor): An array of shape BxCxHxW, where C is the number
+            of masks and H=W=256. These low res logits can be passed to
+            a subsequent iteration as mask input.
+        """
+        if not self.is_image_set:
+            raise RuntimeError("An image must be set with .set_image(...) before mask prediction.")
+
+        if point_coords is not None:
+            points = (point_coords.to(self.load_device), point_labels.to(self.load_device))
+        else:
+            points = None
+
+        # load
+        if boxes is not None:
+            boxes = boxes.to(self.load_device)
+        if mask_input is not None:
+            mask_input = mask_input.to(self.load_device)
+        model_management.load_model_gpu(self.patcher)
+
+        # Embed prompts
+        sparse_embeddings, dense_embeddings = self.patcher.model.prompt_encoder(
+            points=points,
+            boxes=boxes,
+            masks=mask_input,
+        )
+
+        # Predict masks
+        low_res_masks, iou_predictions = self.patcher.model.mask_decoder(
+            image_embeddings=self.features,
+            image_pe=self.patcher.model.prompt_encoder.get_dense_pe(),
+            sparse_prompt_embeddings=sparse_embeddings,
+            dense_prompt_embeddings=dense_embeddings,
+            multimask_output=multimask_output,
+        )
+
+        # Upscale the masks to the original image resolution
+        masks = self.patcher.model.postprocess_masks(low_res_masks, self.input_size, self.original_size)
+
+        if not return_logits:
+            masks = masks > self.patcher.model.mask_threshold
+
+        return masks, iou_predictions, low_res_masks
+
+    def get_image_embedding(self) -> torch.Tensor:
+        """
+        Returns the image embeddings for the currently set image, with
+        shape 1xCxHxW, where C is the embedding dimension and (H,W) are
+        the embedding spatial dimension of SAM (typically C=256, H=W=64).
+        """
+        if not self.is_image_set:
+            raise RuntimeError(
+                "An image must be set with .set_image(...) to generate an embedding."
+            )
+        assert self.features is not None, "Features must exist if an image has been set."
+        return self.features
+
+    @property
+    def device(self) -> torch.device:
+        return self.patcher.model.device
+
+    def reset_image(self) -> None:
+        """Resets the currently set image."""
+        self.is_image_set = False
+        self.features = None
+        self.orig_h = None
+        self.orig_w = None
+        self.input_h = None
+        self.input_w = None