Add OTX optimize for visual prompting task

CHANGELOG.md

@@ -15,6 +15,7 @@ All notable changes to this project will be documented in this file.
 - Add new visual prompting task: train/eval (https://github.com/openvinotoolkit/training_extensions/pull/2203)
 - Add new visual prompting task: export (https://github.com/openvinotoolkit/training_extensions/pull/2274)
 - Add new visual prompting task: deploy (https://github.com/openvinotoolkit/training_extensions/pull/2311)
+- Add new visual prompting task: optimize (PTQ) (https://github.com/openvinotoolkit/training_extensions/pull/2318)
 - Add new object detector ResNeXt101-ATSS (<https://github.com/openvinotoolkit/training_extensions/pull/2309>)
 
 ### Enhancements

src/otx/algorithms/visual_prompting/adapters/openvino/model_wrappers/openvino_models.py

@@ -63,14 +63,16 @@ def __init__(
         preload: bool = False,
     ):
         super().__init__(model_adapter, configuration, preload)
-        self.output_blob_name = "low_res_masks"
 
     @classmethod
     def parameters(cls):  # noqa: D102
         parameters = super().parameters()
         parameters.update({"image_size": NumericalValue(value_type=int, default_value=1024, min=0, max=2048)})
         return parameters
 
+    def _get_outputs(self):
+        return "low_res_masks"
+
     def preprocess(self, inputs: Dict[str, Any], meta: Dict[str, Any]):
         """Preprocess prompts."""
         processed_prompts = []

src/otx/algorithms/visual_prompting/adapters/pytorch_lightning/config/visual_prompting_config.py

@@ -97,8 +97,8 @@ def update_visual_prompting_config(
     if groups:
         for group in groups:
             if group in ["learning_parameters", "nncf_optimization", "pot_parameters", "postprocessing"]:
-                if group in ["nncf_optimization", "pot_parameters"]:
-                    # TODO (sungchul): Consider pot_parameters, nncf_optimization, and postprocessing
+                if group in ["nncf_optimization"]:
+                    # TODO (sungchul): Consider nncf_optimization
                     logger.warning(f"{group} will be implemented.")
                     continue
                 update_visual_prompting_config(visual_prompting_config, getattr(otx_config, group))

src/otx/algorithms/visual_prompting/adapters/pytorch_lightning/datasets/pipelines/sam_transforms.py

@@ -10,7 +10,6 @@
 import numpy as np
 import torch
 from torch import Tensor
-from torch.nn import functional as F
 from torchvision.transforms.functional import resize, to_pil_image  # type: ignore
 
 
@@ -36,24 +35,26 @@ def __call__(self, item: Dict[str, Union[List, Tensor]]) -> Dict[str, Union[List
         Dict[str, Union[List, Tensor]]: Dictionary of batch data.
         """
         item["images"] = torch.as_tensor(
-            self.apply_image(item["images"]).transpose((2, 0, 1)), dtype=torch.get_default_dtype()
+            self.apply_image(item["images"], self.target_length).transpose((2, 0, 1)), dtype=torch.get_default_dtype()
         )
         item["gt_masks"] = [torch.as_tensor(gt_mask) for gt_mask in item["gt_masks"]]
         item["bboxes"] = self.apply_boxes(item["bboxes"], item["original_size"])
         if item["points"]:
             item["points"] = self.apply_coords(item["points"], item["original_size"])
         return item
 
-    def apply_image(self, image: np.ndarray) -> np.ndarray:
+    @classmethod
+    def apply_image(cls, image: np.ndarray, target_length: int) -> np.ndarray:
         """Expects a numpy array with shape HxWxC in uint8 format.
 
         Args:
             image (np.ndarray): Image array.
+            target_length (int): The length of the longest side of the image.
 
         Returns:
             np.ndarray: Resized image.
         """
-        target_size = self.get_preprocess_shape(image.shape[0], image.shape[1], self.target_length)
+        target_size = cls.get_preprocess_shape(image.shape[0], image.shape[1], target_length)
         return np.array(resize(to_pil_image(image), target_size))
 
     def apply_coords(self, coords: np.ndarray, original_size: Union[List[Any], Tensor]) -> np.ndarray:
@@ -88,56 +89,6 @@ def apply_boxes(self, boxes: np.ndarray, original_size: Union[List[Any], Tensor]
         boxes = self.apply_coords(boxes.reshape(-1, 2, 2), original_size)
         return boxes.reshape(-1, 4)
 
-    def apply_image_torch(self, image: torch.Tensor) -> torch.Tensor:
-        """Expects batched images with shape BxCxHxW and float format.
-
-        This transformation may not exactly match apply_image.
-        apply_image is the transformation expected by the model.
-
-        Args:
-            image (torch.Tensor): Image tensor.
-
-        Returns:
-            torch.Tensor: Resized image.
-        """
-        # Expects an image in BCHW format. May not exactly match apply_image.
-        target_size = self.get_preprocess_shape(image.shape[2], image.shape[3], self.target_length)
-        return F.interpolate(image, target_size, mode="bilinear", align_corners=False, antialias=True)
-
-    def apply_coords_torch(self, coords: torch.Tensor, original_size: Tuple[int, ...]) -> torch.Tensor:
-        """Expects a torch tensor with length 2 in the last dimension.
-
-        Requires the original image size in (H, W) format.
-
-        Args:
-            coords (torch.Tensor): Coordinates tensor.
-            original_size (Tuple[int, ...]): Original size of image.
-
-        Returns:
-            torch.Tensor: Resized coordinates.
-        """
-        old_h, old_w = original_size
-        new_h, new_w = self.get_preprocess_shape(original_size[0], original_size[1], self.target_length)
-        coords = deepcopy(coords).to(torch.float)
-        coords[..., 0] = coords[..., 0] * (new_w / old_w)
-        coords[..., 1] = coords[..., 1] * (new_h / old_h)
-        return coords
-
-    def apply_boxes_torch(self, boxes: torch.Tensor, original_size: Tuple[int, ...]) -> torch.Tensor:
-        """Expects a torch tensor with shape Bx4.
-
-        Requires the original image size in (H, W) format.
-
-        Args:
-            boxes (torch.Tensor): Boxes tensor.
-            original_size (Tuple[int, ...]): Original size of image.
-
-        Returns:
-            torch.Tensor: Resized boxes.
-        """
-        boxes = self.apply_coords_torch(boxes.reshape(-1, 2, 2), original_size)
-        return boxes.reshape(-1, 4)
-
     @staticmethod
     def get_preprocess_shape(oldh: int, oldw: int, long_side_length: int) -> Tuple[int, int]:
         """Compute the output size given input size and target long side length.

src/otx/algorithms/visual_prompting/adapters/pytorch_lightning/models/visual_prompters/segment_anything.py

@@ -174,9 +174,9 @@ def replace_state_dict_keys(state_dict, revise_keys):
                 state_dict = replace_state_dict_keys(state_dict, revise_keys)
                 self.load_state_dict(state_dict)
 
-    #################################################
-    #     forward for inference (export/deploy)     #
-    #################################################
+    ##########################################################
+    #     forward for inference (export/deploy/optimize)     #
+    ##########################################################
     @torch.no_grad()
     def forward(
         self,
@@ -185,7 +185,7 @@ def forward(
         point_labels: Tensor,
         mask_input: Tensor,
         has_mask_input: Tensor,
-        orig_size: Tensor,
+        # orig_size: Tensor,
     ):
         """Forward method for SAM inference (export/deploy).
 
@@ -227,16 +227,18 @@ def forward(
         if self.config.model.return_single_mask:
             masks, scores = self.select_masks(masks, scores, point_coords.shape[1])
 
-        upscaled_masks = self.mask_postprocessing(masks, orig_size[0])
+        return scores, masks
+        # TODO (sungchul): apply inner postprocessing
+        # upscaled_masks = self.mask_postprocessing(masks, orig_size[0])
 
-        if self.config.model.return_extra_metrics:
-            stability_scores = self.calculate_stability_score(
-                upscaled_masks, self.config.model.mask_threshold, self.config.model.stability_score_offset
-            )
-            areas = (upscaled_masks > self.config.model.mask_threshold).sum(-1).sum(-1)
-            return upscaled_masks, scores, stability_scores, areas, masks
+        # if self.config.model.return_extra_metrics:
+        #     stability_scores = self.calculate_stability_score(
+        #         upscaled_masks, self.config.model.mask_threshold, self.config.model.stability_score_offset
+        #     )
+        #     areas = (upscaled_masks > self.config.model.mask_threshold).sum(-1).sum(-1)
+        #     return upscaled_masks, scores, stability_scores, areas, masks
 
-        return upscaled_masks, scores, masks
+        # return upscaled_masks, scores, masks
 
     def _embed_points(self, point_coords: Tensor, point_labels: Tensor) -> Tensor:
         """Embed sparse input prompts.

src/otx/algorithms/visual_prompting/configs/base/configuration.py

@@ -17,13 +17,15 @@
 
 from attr import attrs
 
-from otx.algorithms.common.configs import BaseConfig
+from otx.algorithms.common.configs import BaseConfig, POTQuantizationPreset
 from otx.api.configuration.elements import (
     ParameterGroup,
     add_parameter_group,
+    boolean_attribute,
     configurable_boolean,
     configurable_float,
     configurable_integer,
+    selectable,
     string_attribute,
 )
 from otx.api.configuration.model_lifecycle import ModelLifecycle
@@ -95,5 +97,20 @@ class __Postprocessing(ParameterGroup):
             affects_outcome_of=ModelLifecycle.INFERENCE,
         )
 
+    @attrs
+    class __POTParameter(BaseConfig.BasePOTParameter):
+        header = string_attribute("POT Parameters")
+        description = header
+        visible_in_ui = boolean_attribute(False)
+
+        preset = selectable(
+            default_value=POTQuantizationPreset.MIXED,
+            header="Preset",
+            description="Quantization preset that defines quantization scheme",
+            editable=True,
+            visible_in_ui=True,
+        )
+
     learning_parameters = add_parameter_group(__LearningParameters)
     postprocessing = add_parameter_group(__Postprocessing)
+    pot_parameters = add_parameter_group(__POTParameter)

src/otx/algorithms/visual_prompting/configs/configuration.yaml

@@ -148,7 +148,7 @@ pot_parameters:
     affects_outcome_of: NONE
     auto_hpo_state: not_possible
     auto_hpo_value: null
-    default_value: Performance
+    default_value: Mixed
     description: Quantization preset that defines quantization scheme
     editable: true
     enum_name: POTQuantizationPreset
@@ -162,7 +162,7 @@ pot_parameters:
       operator: AND
       rules: []
       type: UI_RULES
-    value: Performance
+    value: Mixed
     visible_in_ui: true
     warning: null
   stat_subset_size:

src/otx/algorithms/visual_prompting/configs/sam_vit_b/config.yaml

@@ -1,6 +1,6 @@
 dataset:
   task: visual_prompting
-  train_batch_size: 2
+  train_batch_size: 4
   val_batch_size: 1
   test_batch_size: 1
   num_workers: 4
@@ -35,7 +35,7 @@ model:
 
 optimizer:
   name: Adam
-  lr: 0.0001
+  lr: 0.000001
 
 callback:
   checkpoint: # arguments for ModelCheckpoint

src/otx/algorithms/visual_prompting/configs/sam_vit_b/configuration.yaml

@@ -85,70 +85,14 @@ learning_parameters:
       visible_in_ui: true
       warning: null
       auto_hpo_state: NOT_POSSIBLE
-nncf_optimization:
-  description: Optimization by NNCF
-  enable_pruning:
-    affects_outcome_of: NONE
-    auto_hpo_state: not_possible
-    auto_hpo_value: null
-    default_value: false
-    description: Enable filter pruning algorithm
-    editable: true
-    header: Enable filter pruning algorithm
-    type: BOOLEAN
-    ui_rules:
-      action: DISABLE_EDITING
-      operator: AND
-      rules: []
-      type: UI_RULES
-    value: false
-    visible_in_ui: true
-    warning: null
-  enable_quantization:
-    affects_outcome_of: NONE
-    auto_hpo_state: not_possible
-    auto_hpo_value: null
-    default_value: true
-    description: Enable quantization algorithm
-    editable: true
-    header: Enable quantization algorithm
-    type: BOOLEAN
-    ui_rules:
-      action: DISABLE_EDITING
-      operator: AND
-      rules: []
-      type: UI_RULES
-    value: true
-    visible_in_ui: true
-    warning: null
-  header: Optimization by NNCF
-  pruning_supported:
-    affects_outcome_of: TRAINING
-    auto_hpo_state: not_possible
-    auto_hpo_value: null
-    default_value: false
-    description: Whether filter pruning is supported
-    editable: false
-    header: Whether filter pruning is supported
-    type: BOOLEAN
-    ui_rules:
-      action: DISABLE_EDITING
-      operator: AND
-      rules: []
-      type: UI_RULES
-    value: false
-    visible_in_ui: false
-    warning: null
-  type: PARAMETER_GROUP
-  visible_in_ui: true
 pot_parameters:
   description: POT Parameters
   header: POT Parameters
   preset:
     affects_outcome_of: NONE
     auto_hpo_state: not_possible
     auto_hpo_value: null
-    default_value: Performance
+    default_value: Mixed
     description: Quantization preset that defines quantization scheme
     editable: true
     enum_name: POTQuantizationPreset
@@ -162,7 +106,7 @@ pot_parameters:
       operator: AND
       rules: []
       type: UI_RULES
-    value: Performance
+    value: Mixed
     visible_in_ui: true
     warning: null
   stat_subset_size:

src/otx/algorithms/visual_prompting/tasks/inference.py

@@ -281,9 +281,8 @@ def _export_to_onnx(self, onnx_path: Dict[str, str]):
                     "point_labels": torch.randint(low=0, high=4, size=(1, 2), dtype=torch.float),
                     "mask_input": torch.randn(1, 1, *mask_input_size, dtype=torch.float),
                     "has_mask_input": torch.tensor([[1]], dtype=torch.float),
-                    "orig_size": torch.tensor([[height, width]], dtype=torch.float),
                 }
-                output_names = ["masks", "iou_predictions", "low_res_masks"]
+                output_names = ["iou_predictions", "low_res_masks"]
                 model_to_export = self.model
 
             with warnings.catch_warnings():