merge develop

.ci/requirements/publish/requirements.txt

@@ -7,7 +7,7 @@
 build==1.1.1 \
     --hash=sha256:8ed0851ee76e6e38adce47e4bee3b51c771d86c64cf578d0c2245567ee200e73 \
     --hash=sha256:8eea65bb45b1aac2e734ba2cc8dad3a6d97d97901a395bd0ed3e7b46953d2a31
-    # via -r .ci/publish-deps.in
+    # via -r requirements.in
 certifi==2024.2.2 \
     --hash=sha256:0569859f95fc761b18b45ef421b1290a0f65f147e92a1e5eb3e635f9a5e4e66f \
     --hash=sha256:dc383c07b76109f368f6106eee2b593b04a011ea4d55f652c6ca24a754d1cdd1
@@ -196,9 +196,9 @@ docutils==0.20.1 \
     --hash=sha256:96f387a2c5562db4476f09f13bbab2192e764cac08ebbf3a34a95d9b1e4a59d6 \
     --hash=sha256:f08a4e276c3a1583a86dce3e34aba3fe04d02bba2dd51ed16106244e8a923e3b
     # via readme-renderer
-idna==3.6 \
-    --hash=sha256:9ecdbbd083b06798ae1e86adcbfe8ab1479cf864e4ee30fe4e46a003d12491ca \
-    --hash=sha256:c05567e9c24a6b9faaa835c4821bad0590fbb9d5779e7caa6e1cc4978e7eb24f
+idna==3.7 \
+    --hash=sha256:028ff3aadf0609c1fd278d8ea3089299412a7a8b9bd005dd08b9f8285bcb5cfc \
+    --hash=sha256:82fee1fc78add43492d3a1898bfa6d8a904cc97d8427f683ed8e798d07761aa0
     # via requests
 importlib-metadata==7.0.2 \
     --hash=sha256:198f568f3230878cb1b44fbd7975f87906c22336dba2e4a7f05278c281fbd792 \
@@ -307,7 +307,7 @@ tomli==2.0.1 \
 twine==5.0.0 \
     --hash=sha256:89b0cc7d370a4b66421cc6102f269aa910fe0f1861c124f573cf2ddedbc10cf4 \
     --hash=sha256:a262933de0b484c53408f9edae2e7821c1c45a3314ff2df9bdd343aa7ab8edc0
-    # via -r .ci/publish-deps.in
+    # via -r requirements.in
 urllib3==2.2.1 \
     --hash=sha256:450b20ec296a467077128bff42b73080516e71b56ff59a60a02bef2232c4fa9d \
     --hash=sha256:d0570876c61ab9e520d776c38acbbb5b05a776d3f9ff98a5c8fd5162a444cf19

.dockerignore

@@ -5,3 +5,4 @@
 !LICENSE
 !MANIFEST.in
 !docker/download_pretrained_weights.py
+!.ci

CHANGELOG.md

@@ -8,6 +8,54 @@ All notable changes to this project will be documented in this file.
 
 ### Enhancements
 
+## \[v1.6.1\]
+
+### Enhancements
+
+- Update pymongo version to 4.6.3 for resolving CVE-2024-21506
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3396>)
+- Use torchvision in MRCNN on CUDA
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3347>)
+- Update IPEX version in installation guide documentation
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3343>)
+- Update benchmark
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3338>)
+- Bump idan version to 3.7
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3332>)
+- Support benchmark history summary
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3307>)
+- Pin pymongo version to 4.5.0
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3316>)
+- Upgrade MAPI
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3304>)
+- Add NMS iou threshold configurable parameter
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3287>)
+- Remedy some medium/low severity bandit issues
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3208>)
+- Update documentations
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3280>)
+- Add perf benchmark test cases for action and visual prompting
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3292>)
+
+### Bug fixes
+
+- Explicitly cast incorrect output type in OV model
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3395>)
+- Update QAT configs for rotated detection
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3375>)
+- Hotfix :wrench: Bypass ClsIncrSampler for tiling
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3374>)
+- [NNCF] Dynamic shape datasets WA
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3355>)
+- [Hotfix] :fire: Fixing detection oriented OV inferencer
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3351>)
+- Revert adaptive batch size
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3340>)
+- Fix e2e tests for XPU
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3305>)
+- Remove torch.xpu.optimize for semantic_segmentation task
+  (<https://github.com/openvinotoolkit/training_extensions/pull/3172>)
+
 ## \[1.6.0\]
 
 ### New features

docs/source/guide/release_notes/index.rst

@@ -5,9 +5,37 @@ Releases
   :maxdepth: 1
 
 
-v2.0.0 (1Q24)
+v2.0.0 (2Q24)
 -------------
 
+v1.6.1 (2024.05)
+----------------
+
+Enhancements
+^^^^^^^^^^^^
+- Update pymongo version to 4.6.3 for resolving CVE-2024-21506
+- Use torchvision in MRCNN on CUDA
+- Update IPEX version in installation guide documentation
+- Update benchmark
+- Bump idan version to 3.7
+- Support benchmark history summary
+- Upgrade MAPI
+- Add NMS iou threshold configurable parameter
+- Remedy some medium/low severity bandit issues
+- Update documentations
+- Add perf benchmark test cases for action and visual prompting
+
+Bug fixes
+^^^^^^^^^
+- Explicitly cast incorrect output type in OV model
+- Update QAT configs for rotated detection
+- Hotfix :wrench: Bypass ClsIncrSampler for tiling
+- [NNCF] Dynamic shape datasets WA
+- [Hotfix] :fire: Fixing detection oriented OV inferencer
+- Revert adaptive batch size
+- Fix e2e tests for XPU
+- Remove torch.xpu.optimize for semantic_segmentation task
+
 v1.6.0 (2024.04)
 ----------------
 
@@ -59,7 +87,6 @@ v1.5.2 (2024.01)
 
 Enhancements
 ^^^^^^^^^^^^
-
 - Add memory bounded datumaro data format detect
 - Remove Protobuf version limitation (<4)
 
@@ -155,6 +182,7 @@ v1.4.0 (3Q23)
 
 v1.3.1 (2Q23)
 -------------
+
 - Minor bug fixes
 
 v1.3.0 (2Q23)
@@ -176,6 +204,7 @@ v1.3.0 (2Q23)
 
 v1.2.4 (3Q23)
 -------------
+
 - Per-class saliency maps for M-RCNN
 - Disable semantic segmentation soft prediction processing
 - Update export and nncf hyperparameters

pyproject.toml

@@ -334,6 +334,7 @@ max-returns = 10
     "ANN001",   # Skip annotation type hint in test codes
     "ANN201",   # Skip return type hint in test codes
     "D",     # Test skips missing docstring argument with magic (fixture) methods.
+    "ARG001",   # Some arguments are passed for executing fixture
 ]
 "src/otx/**/*.py" = [
     "ERA001",

src/otx/__init__.py

@@ -3,7 +3,7 @@
 # Copyright (C) 2023 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 
-__version__ = "2.1.0rc0"
+__version__ = "2.0.0rc0"
 
 from otx.core.types import *  # noqa: F403
 

src/otx/algo/accelerators/xpu.py

@@ -4,17 +4,12 @@
 #
 from __future__ import annotations
 
-from typing import Any, Union
+from typing import Any
 
-import numpy as np
 import torch
 from lightning.pytorch.accelerators import AcceleratorRegistry
 from lightning.pytorch.accelerators.accelerator import Accelerator
-from mmcv.ops.nms import NMSop
-from mmcv.ops.roi_align import RoIAlign
-from mmengine.structures import instance_data
 
-from otx.algo.detection.utils import monkey_patched_nms, monkey_patched_roi_align
 from otx.utils.utils import is_xpu_available
 
 
@@ -30,7 +25,6 @@ def setup_device(self, device: torch.device) -> None:
             raise RuntimeError(msg)
 
         torch.xpu.set_device(device)
-        self.patch_packages_xpu()
 
     @staticmethod
     def parse_devices(devices: str | list | torch.device) -> list:
@@ -59,26 +53,7 @@ def get_device_stats(self, device: str | torch.device) -> dict[str, Any]:
         return {}
 
     def teardown(self) -> None:
-        """Cleans-up XPU-related resources."""
-        self.revert_packages_xpu()
-
-    def patch_packages_xpu(self) -> None:
-        """Patch packages when xpu is available."""
-        # patch instance_data from mmengie
-        long_type_tensor = Union[torch.LongTensor, torch.xpu.LongTensor]
-        bool_type_tensor = Union[torch.BoolTensor, torch.xpu.BoolTensor]
-        instance_data.IndexType = Union[str, slice, int, list, long_type_tensor, bool_type_tensor, np.ndarray]
-
-        # patch nms and roi_align
-        self._nms_op_forward = NMSop.forward
-        self._roi_align_forward = RoIAlign.forward
-        NMSop.forward = monkey_patched_nms
-        RoIAlign.forward = monkey_patched_roi_align
-
-    def revert_packages_xpu(self) -> None:
-        """Revert packages when xpu is available."""
-        NMSop.forward = self._nms_op_forward
-        RoIAlign.forward = self._roi_align_forward
+        """Clean up any state created by the accelerator."""
 
 
 AcceleratorRegistry.register(

src/otx/algo/callbacks/adaptive_train_scheduling.py

@@ -29,11 +29,17 @@ class AdaptiveTrainScheduling(Callback):
             Defaults to -0.025.
     """
 
-    def __init__(self, max_interval: int = 5, decay: float = -0.025):
+    def __init__(
+        self,
+        max_interval: int = 5,
+        decay: float = -0.025,
+        min_earlystop_patience: int = 3,
+        min_lrschedule_patience: int = 2,
+    ):
         self.max_interval = max_interval
         self.decay = decay
-        self.min_earlystop_interval = 3
-        self.min_lrschedule_patience = 2
+        self.min_earlystop_patience = min_earlystop_patience
+        self.min_lrschedule_patience = min_lrschedule_patience
         self._saved_check_val_every_n_epoch: int | None = None
         self._saved_log_every_n_steps: int | None = None
         self._revert_lr_frequency: list = []
@@ -153,7 +159,7 @@ def _revert_func(callback: Callback, saved_patience: int) -> None:
 
         for callback in callbacks:
             if isinstance(callback, EarlyStopping):
-                adjusted_patience = max(int(callback.patience / adaptive_interval), self.min_earlystop_interval)
+                adjusted_patience = max(int(callback.patience / adaptive_interval), self.min_earlystop_patience)
                 msg = (
                     "The patience of early stopping will be changed due to the effect of adaptive interval: "
                     f"{callback.patience} --> {adjusted_patience}."

src/otx/algo/classification/classifier/base_classifier.py

@@ -177,6 +177,7 @@ def predict(self, inputs: torch.Tensor, **kwargs) -> list[torch.Tensor]:
         feats = self.extract_feat(inputs)
         return self.head.predict(feats, **kwargs)
 
+    @torch.no_grad()
     def _forward_explain(self, images: torch.Tensor) -> dict[str, torch.Tensor | list[torch.Tensor]]:
         """Generates explanations for the given images using the classifier.
 
@@ -186,8 +187,8 @@ def _forward_explain(self, images: torch.Tensor) -> dict[str, torch.Tensor | lis
         Returns:
             dict[str, torch.Tensor | list[torch.Tensor]]: A dictionary containing the following keys:
                 - "logits" (torch.Tensor): The output logits from the classifier.
-                - "preds" (torch.Tensor): The predicted class labels.
-                - "scores" (torch.Tensor): The softmax scores for each class.
+                - "preds" (torch.Tensor): The predicted class labels. Only included in non-tracing mode.
+                - "scores" (torch.Tensor): The softmax scores for each class. Only included in non-tracing mode.
                 - "saliency_map" (torch.Tensor): The saliency map generated by the explainer.
                 - "feature_vector" (torch.Tensor): The feature vector extracted from the backbone network.
         """
@@ -212,14 +213,18 @@ def _forward_explain(self, images: torch.Tensor) -> dict[str, torch.Tensor | lis
             scores = pred_results.unbind(0)
             preds = logits.argmax(-1, keepdim=True).unbind(0)
 
-        return {
+        outputs = {
             "logits": logits,
             "feature_vector": feature_vector,
             "saliency_map": saliency_map,
-            "scores": scores,
-            "preds": preds,
         }
 
+        if not torch.jit.is_tracing():
+            outputs["scores"] = scores
+            outputs["preds"] = preds
+
+        return outputs
+
     @torch.no_grad()
     def _head_forward_fn(self, x: torch.Tensor) -> torch.Tensor:
         """Performs model's neck and head forward."""

src/otx/algo/classification/efficientnet.py

@@ -131,7 +131,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )
@@ -246,7 +246,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )
@@ -392,7 +392,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )

src/otx/algo/classification/efficientnet_v2.py

@@ -128,7 +128,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )
@@ -241,7 +241,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )
@@ -384,7 +384,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )

src/otx/algo/classification/mobilenet_v3.py

@@ -140,7 +140,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )
@@ -257,7 +257,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )
@@ -403,7 +403,7 @@ def _exporter(self) -> OTXModelExporter:
             resize_mode="standard",
             pad_value=0,
             swap_rgb=False,
-            via_onnx=True,  # NOTE: This should be done via onnx
+            via_onnx=False,
             onnx_export_configuration=None,
             output_names=["logits", "feature_vector", "saliency_map"] if self.explain_mode else None,
         )