Add ResNeXt101-ATSS as a Experimental model (#2309)

* Add ResNeXt101-ATSS

* Change atss name to mobilenetv2-atss

* Change to experimental yaml

* Revert mobilnet atss template changes

* Update CHANGELOG.md

* Add pretrained weights url

* Update resnext101_atss stats

* Fix model_api issue

* Fix wrong import in unit tests

Author: jaegukhyun

CHANGELOG.md

@@ -11,8 +11,9 @@ All notable changes to this project will be documented in this file.
 - Auto train type detection for Semi-SL, Self-SL and Incremental: "--train-type" now is optional (https://github.com/openvinotoolkit/training_extensions/pull/2195)
 - Add per-class XAI saliency maps for Mask R-CNN model (https://github.com/openvinotoolkit/training_extensions/pull/2227)
 - Add new object detector Deformable DETR (<https://github.com/openvinotoolkit/training_extensions/pull/2249>)
-- Add new object detecotr DINO(<https://github.com/openvinotoolkit/training_extensions/pull/2266>)
+- Add new object detector DINO(<https://github.com/openvinotoolkit/training_extensions/pull/2266>)
 - Add new visual prompting task (https://github.com/openvinotoolkit/training_extensions/pull/2203), (https://github.com/openvinotoolkit/training_extensions/pull/2274)
+- Add new object detector ResNeXt101-ATSS (<https://github.com/openvinotoolkit/training_extensions/pull/2309>)
 
 ### Enhancements
 

src/otx/algorithms/detection/configs/detection/resnext101_atss/__init__.py

@@ -0,0 +1,4 @@
+"""Initialization of ResNeXt101-ATSS model for Detection Task."""
+
+# Copyright (C) 2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0

src/otx/algorithms/detection/configs/detection/resnext101_atss/compression_config.json

@@ -0,0 +1,74 @@
+{
+  "base": {
+    "find_unused_parameters": true,
+    "nncf_config": {
+      "target_metric_name": "mAP",
+      "input_info": {
+        "sample_size": [1, 3, 992, 736]
+      },
+      "compression": [],
+      "log_dir": "/tmp"
+    }
+  },
+  "nncf_quantization": {
+    "optimizer": {
+      "lr": 0.0005
+    },
+    "nncf_config": {
+      "compression": [
+        {
+          "algorithm": "quantization",
+          "initializer": {
+            "range": {
+              "num_init_samples": 300
+            },
+            "batchnorm_adaptation": {
+              "num_bn_adaptation_samples": 300
+            }
+          }
+        }
+      ],
+      "accuracy_aware_training": {
+        "mode": "early_exit",
+        "params": {
+          "maximal_absolute_accuracy_degradation": 0.01,
+          "maximal_total_epochs": 20
+        }
+      }
+    }
+  },
+  "nncf_quantization_pruning": {
+    "nncf_config": {
+      "accuracy_aware_training": {
+        "mode": "adaptive_compression_level",
+        "params": {
+          "initial_training_phase_epochs": 5,
+          "maximal_total_epochs": 100,
+          "patience_epochs": 5
+        }
+      },
+      "compression": [
+        {
+          "algorithm": "filter_pruning",
+          "params": {
+            "schedule": "baseline",
+            "pruning_flops_target": 0.1,
+            "filter_importance": "geometric_median"
+          }
+        },
+        {
+          "algorithm": "quantization",
+          "initializer": {
+            "range": {
+              "num_init_samples": 300
+            },
+            "batchnorm_adaptation": {
+              "num_bn_adaptation_samples": 300
+            }
+          }
+        }
+      ]
+    }
+  },
+  "order_of_parts": ["nncf_quantization", "nncf_quantization_pruning"]
+}

src/otx/algorithms/detection/configs/detection/resnext101_atss/data_pipeline.py

@@ -0,0 +1,72 @@
+"""Data Pipeline of ResNeXt101-ATSS model for Detection Task."""
+
+# Copyright (C) 2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+# pylint: disable=invalid-name
+
+__img_size = (992, 736)
+__img_norm_cfg = dict(mean=[0, 0, 0], std=[255, 255, 255], to_rgb=True)
+
+train_pipeline = [
+    dict(type="LoadImageFromFile"),
+    dict(type="LoadAnnotations", with_bbox=True),
+    dict(type="MinIoURandomCrop", min_ious=(0.1, 0.3, 0.5, 0.7, 0.9), min_crop_size=0.3),
+    dict(
+        type="Resize",
+        img_scale=[(992, 736), (896, 736), (1088, 736), (992, 672), (992, 800)],
+        multiscale_mode="value",
+        keep_ratio=False,
+    ),
+    dict(type="RandomFlip", flip_ratio=0.5),
+    dict(type="Normalize", **__img_norm_cfg),
+    dict(type="DefaultFormatBundle"),
+    dict(type="Collect", keys=["img", "gt_bboxes", "gt_labels"]),
+]
+test_pipeline = [
+    dict(type="LoadImageFromFile"),
+    dict(
+        type="MultiScaleFlipAug",
+        img_scale=__img_size,
+        flip=False,
+        transforms=[
+            dict(type="Resize", keep_ratio=False),
+            dict(type="RandomFlip"),
+            dict(type="Normalize", **__img_norm_cfg),
+            dict(type="ImageToTensor", keys=["img"]),
+            dict(type="Collect", keys=["img"]),
+        ],
+    ),
+]
+
+
+__dataset_type = "CocoDataset"
+__data_root = "data/coco/"
+
+
+__samples_per_gpu = 2
+
+data = dict(
+    samples_per_gpu=__samples_per_gpu,
+    workers_per_gpu=2,
+    train=dict(
+        type=__dataset_type,
+        ann_file=__data_root + "annotations/instances_train2017.json",
+        img_prefix=__data_root + "train2017/",
+        pipeline=train_pipeline,
+    ),
+    val=dict(
+        type=__dataset_type,
+        ann_file=__data_root + "annotations/instances_val2017.json",
+        img_prefix=__data_root + "val2017/",
+        test_mode=True,
+        pipeline=test_pipeline,
+    ),
+    test=dict(
+        type=__dataset_type,
+        ann_file=__data_root + "annotations/instances_val2017.json",
+        img_prefix=__data_root + "val2017/",
+        test_mode=True,
+        pipeline=test_pipeline,
+    ),
+)

src/otx/algorithms/detection/configs/detection/resnext101_atss/deployment.py

@@ -0,0 +1,14 @@
+"""MMDeploy config of ResNeXt101-ATSS model for Detection Task."""
+
+# Copyright (C) 2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+_base_ = ["../../base/deployments/base_detection_dynamic.py"]
+
+ir_config = dict(
+    output_names=["boxes", "labels"],
+)
+
+backend_config = dict(
+    model_inputs=[dict(opt_shapes=dict(input=[-1, 3, 736, 992]))],
+)

src/otx/algorithms/detection/configs/detection/resnext101_atss/hpo_config.yaml

@@ -0,0 +1,16 @@
+metric: mAP
+search_algorithm: asha
+early_stop: None
+hp_space:
+  learning_parameters.learning_rate:
+    param_type: qloguniform
+    range:
+      - 0.0004
+      - 0.04
+      - 0.0001
+  learning_parameters.batch_size:
+    param_type: qloguniform
+    range:
+      - 4
+      - 16
+      - 2

src/otx/algorithms/detection/configs/detection/resnext101_atss/model.py

@@ -0,0 +1,83 @@
+"""Model configuration of ResNeXt101-ATSS model for Detection Task."""
+
+# Copyright (C) 2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+# pylint: disable=invalid-name
+
+_base_ = [
+    "../../../../../recipes/stages/detection/incremental.py",
+    "../../base/models/detector.py",
+]
+
+model = dict(
+    type="CustomATSS",
+    backbone=dict(
+        type="ResNeXt",
+        depth=101,
+        groups=64,
+        base_width=4,
+        num_stages=4,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=1,
+        norm_cfg=dict(type="BN", requires_grad=True),
+        style="pytorch",
+        init_cfg=dict(type="Pretrained", checkpoint="open-mmlab://resnext101_64x4d"),
+    ),
+    neck=dict(
+        type="FPN",
+        in_channels=[256, 512, 1024, 2048],
+        out_channels=256,
+        start_level=1,
+        add_extra_convs="on_output",
+        num_outs=5,
+        relu_before_extra_convs=True,
+    ),
+    bbox_head=dict(
+        type="CustomATSSHead",
+        num_classes=2,
+        in_channels=256,
+        stacked_convs=4,
+        feat_channels=256,
+        anchor_generator=dict(
+            type="AnchorGenerator",
+            ratios=[1.0],
+            octave_base_scale=8,
+            scales_per_octave=1,
+            strides=[8, 16, 32, 64, 128],
+        ),
+        bbox_coder=dict(
+            type="DeltaXYWHBBoxCoder",
+            target_means=[0.0, 0.0, 0.0, 0.0],
+            target_stds=[0.1, 0.1, 0.2, 0.2],
+        ),
+        loss_cls=dict(type="FocalLoss", use_sigmoid=True, gamma=2.0, alpha=0.25, loss_weight=1.0),
+        loss_bbox=dict(type="GIoULoss", loss_weight=2.0),
+        loss_centerness=dict(type="CrossEntropyLoss", use_sigmoid=True, loss_weight=1.0),
+        use_qfl=False,
+        qfl_cfg=dict(
+            type="QualityFocalLoss",
+            use_sigmoid=True,
+            beta=2.0,
+            loss_weight=1.0,
+        ),
+    ),
+    train_cfg=dict(
+        assigner=dict(type="ATSSAssigner", topk=9),
+        allowed_border=-1,
+        pos_weight=-1,
+        debug=False,
+    ),
+    test_cfg=dict(
+        nms_pre=1000,
+        min_bbox_size=0,
+        score_thr=0.05,
+        nms=dict(type="nms", iou_threshold=0.6),
+        max_per_img=100,
+    ),
+)
+
+load_from = "https://storage.openvinotoolkit.org/repositories/openvino_training_extensions/\
+models/object_detection/v2/resnext101_atss_070623.pth"
+
+fp16 = dict(loss_scale=512.0)

src/otx/algorithms/detection/configs/detection/resnext101_atss/semisl/__init__.py

@@ -0,0 +1,4 @@
+"""Initialization of ResNeXt101-ATSS model for Semi-SL Detection Task."""
+
+# Copyright (C) 2023 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0

src/otx/algorithms/detection/configs/detection/resnext101_atss/semisl/compression_config.json

@@ -0,0 +1,74 @@
+{
+  "base": {
+    "find_unused_parameters": true,
+    "nncf_config": {
+      "target_metric_name": "mAP",
+      "input_info": {
+        "sample_size": [1, 3, 992, 736]
+      },
+      "compression": [],
+      "log_dir": "/tmp"
+    }
+  },
+  "nncf_quantization": {
+    "optimizer": {
+      "lr": 0.0005
+    },
+    "nncf_config": {
+      "compression": [
+        {
+          "algorithm": "quantization",
+          "initializer": {
+            "range": {
+              "num_init_samples": 300
+            },
+            "batchnorm_adaptation": {
+              "num_bn_adaptation_samples": 300
+            }
+          }
+        }
+      ],
+      "accuracy_aware_training": {
+        "mode": "early_exit",
+        "params": {
+          "maximal_absolute_accuracy_degradation": 0.01,
+          "maximal_total_epochs": 20
+        }
+      }
+    }
+  },
+  "nncf_quantization_pruning": {
+    "nncf_config": {
+      "accuracy_aware_training": {
+        "mode": "adaptive_compression_level",
+        "params": {
+          "initial_training_phase_epochs": 5,
+          "maximal_total_epochs": 100,
+          "patience_epochs": 5
+        }
+      },
+      "compression": [
+        {
+          "algorithm": "filter_pruning",
+          "params": {
+            "schedule": "baseline",
+            "pruning_flops_target": 0.1,
+            "filter_importance": "geometric_median"
+          }
+        },
+        {
+          "algorithm": "quantization",
+          "initializer": {
+            "range": {
+              "num_init_samples": 300
+            },
+            "batchnorm_adaptation": {
+              "num_bn_adaptation_samples": 300
+            }
+          }
+        }
+      ]
+    }
+  },
+  "order_of_parts": ["nncf_quantization", "nncf_quantization_pruning"]
+}