Merge pull request #985 from openvinotoolkit/ik/move_mmdet_templates

move mmdet templates

Author: Ilya-Krylov

.gitmodules

@@ -1,5 +1,5 @@
 [submodule "external/mmdetection"]
-	path = external/mmdetection
+	path = external/mmdetection/submodule
 	url = ../../openvinotoolkit/mmdetection
 [submodule "external/deep-object-reid"]
 	path = external/deep-object-reid/submodule

.pylintrc

@@ -267,7 +267,7 @@ ignored-classes=optparse.Values,thread._local,_thread._local
 # (useful for modules/projects where namespaces are manipulated during runtime
 # and thus existing member attributes cannot be deduced by static analysis). It
 # supports qualified module names, as well as Unix pattern matching.
-ignored-modules=ote_sdk,mmseg,segmentation_tasks,mmdet,torchreid,torchreid_tasks,cv2,anomalib,pytorch_lightning,torch,addict,compression,openvino,pandas
+ignored-modules=ote_sdk,mmseg,segmentation_tasks,mmdet,detection_tasks,torchreid,torchreid_tasks,cv2,anomalib,pytorch_lightning,torch,addict,compression,openvino,pandas
 
 # Show a hint with possible names when a member name was not found. The aspect
 # of finding the hint is based on edit distance.

external/README.md

@@ -25,21 +25,21 @@ Custom_Image_Classification_EfficientNet-V2-S | EfficientNet-V2-S | 5.76 | 20.23
 ## Object Detection
 ID | Name | Complexity (GFlops) | Model size (MB) | Path
 ------- | ------- | ------- | ------- | -------
-Custom_Object_Detection_YOLOX | YOLOX | 6.5 | 20.4 | mmdetection/configs/ote/custom-object-detection/cspdarknet_YOLOX/template.yaml
-Custom_Object_Detection_Gen3_SSD | SSD | 9.4 | 7.6 | mmdetection/configs/ote/custom-object-detection/gen3_mobilenetV2_SSD/template.yaml
-Custom_Object_Detection_Gen3_ATSS | ATSS | 20.6 | 9.1 | mmdetection/configs/ote/custom-object-detection/gen3_mobilenetV2_ATSS/template.yaml
+Custom_Object_Detection_YOLOX | YOLOX | 6.5 | 20.4 | mmdetection/configs/custom-object-detection/cspdarknet_YOLOX/template.yaml
+Custom_Object_Detection_Gen3_SSD | SSD | 9.4 | 7.6 | mmdetection/configs/custom-object-detection/gen3_mobilenetV2_SSD/template.yaml
+Custom_Object_Detection_Gen3_ATSS | ATSS | 20.6 | 9.1 | mmdetection/configs/custom-object-detection/gen3_mobilenetV2_ATSS/template.yaml
 
 ## Object Counting
 ID | Name | Complexity (GFlops) | Model size (MB) | Path
 ------- | ------- | ------- | ------- | -------
-Custom_Counting_Instance_Segmentation_MaskRCNN_EfficientNetB2B | MaskRCNN-EfficientNetB2B | 68.48 | 13.27 | mmdetection/configs/ote/custom-counting-instance-seg/efficientnetb2b_maskrcnn/template.yaml
-Custom_Counting_Instance_Segmentation_MaskRCNN_ResNet50 | MaskRCNN-ResNet50 | 533.8 | 177.9 | mmdetection/configs/ote/custom-counting-instance-seg/resnet50_maskrcnn/template.yaml
+Custom_Counting_Instance_Segmentation_MaskRCNN_EfficientNetB2B | MaskRCNN-EfficientNetB2B | 68.48 | 13.27 | mmdetection/configs/custom-counting-instance-seg/efficientnetb2b_maskrcnn/template.yaml
+Custom_Counting_Instance_Segmentation_MaskRCNN_ResNet50 | MaskRCNN-ResNet50 | 533.8 | 177.9 | mmdetection/configs/custom-counting-instance-seg/resnet50_maskrcnn/template.yaml
 
 ## Rotated Object Detection
 ID | Name | Complexity (GFlops) | Model size (MB) | Path
 ------- | ------- | ------- | ------- | -------
-Custom_Rotated_Detection_via_Instance_Segmentation_MaskRCNN_EfficientNetB2B | MaskRCNN-EfficientNetB2B | 68.48 | 13.27 | mmdetection/configs/ote/rotated_detection/efficientnetb2b_maskrcnn/template.yaml
-Custom_Rotated_Detection_via_Instance_Segmentation_MaskRCNN_ResNet50 | MaskRCNN-ResNet50 | 533.8 | 177.9 | mmdetection/configs/ote/rotated_detection/resnet50_maskrcnn/template.yaml
+Custom_Rotated_Detection_via_Instance_Segmentation_MaskRCNN_EfficientNetB2B | MaskRCNN-EfficientNetB2B | 68.48 | 13.27 | mmdetection/configs/rotated_detection/efficientnetb2b_maskrcnn/template.yaml
+Custom_Rotated_Detection_via_Instance_Segmentation_MaskRCNN_ResNet50 | MaskRCNN-ResNet50 | 533.8 | 177.9 | mmdetection/configs/rotated_detection/resnet50_maskrcnn/template.yaml
 
 ## Semantic Segmentaion
 ID | Name | Complexity (GFlops) | Model size (MB) | Path

external/mmdetection

@@ -0,0 +1,57 @@
+dataset_type = 'CocoDataset'
+img_size = (1024, 1024)
+
+img_norm_cfg = dict(
+    mean=(103.53, 116.28, 123.675), std=(1.0, 1.0, 1.0), to_rgb=False)
+
+train_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(type='LoadAnnotations', with_bbox=True,
+         with_mask=True, poly2mask=False),
+    dict(type='Resize', img_scale=img_size, keep_ratio=False),
+    dict(type='RandomFlip', flip_ratio=0.5),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='Pad', size_divisor=32),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks'])
+]
+
+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='Pad', size_divisor=32),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img'])
+        ])
+]
+
+data = dict(samples_per_gpu=4,
+            workers_per_gpu=2,
+            train=dict(
+                type='RepeatDataset',
+                adaptive_repeat_times=True,
+                times=1,
+                dataset=dict(
+                    type=dataset_type,
+                    ann_file='data/coco/annotations/instances_train2017.json',
+                    img_prefix='data/coco/train2017',
+                    pipeline=train_pipeline)),
+            val=dict(
+                type=dataset_type,
+                test_mode=True,
+                ann_file='data/coco/annotations/instances_val2017.json',
+                img_prefix='data/coco/val2017',
+                pipeline=test_pipeline),
+            test=dict(
+                type=dataset_type,
+                test_mode=True,
+                ann_file='data/coco/annotations/instances_val2017.json',
+                img_prefix='data/coco/val2017',
+                pipeline=test_pipeline))

external/mmdetection/configs/custom-counting-instance-seg/efficientnetb2b_maskrcnn/coco_data_pipeline.py

@@ -0,0 +1,43 @@
+{
+    "base": {
+        "find_unused_parameters": true,
+        "nncf_config": {
+            "target_metric_name": "mAP",
+            "input_info": {
+                "sample_size": [1, 3, 1024, 1024]
+            },
+            "compression": [],
+            "log_dir": "."
+        }
+    },
+    "nncf_quantization": {
+        "optimizer": {
+            "lr": 0.0005
+        },
+        "nncf_config": {
+            "compression": [
+                {
+                    "algorithm": "quantization",
+                    "initializer": {
+                        "range": {
+                            "num_init_samples": 1000
+                        },
+                        "batchnorm_adaptation": {
+                            "num_bn_adaptation_samples": 1000
+                        }
+                    }
+                }
+            ],
+            "accuracy_aware_training": {
+                "mode": "early_exit",
+                "params": {
+                    "maximal_absolute_accuracy_degradation": 0.01,
+                    "maximal_total_epochs": 20
+                }
+            }
+        }
+    },
+	"order_of_parts": [
+		"nncf_quantization"
+    ]
+}

external/mmdetection/configs/custom-counting-instance-seg/efficientnetb2b_maskrcnn/compression_config.json

@@ -0,0 +1,16 @@
+metric: mAP
+search_algorithm: smbo
+early_stop: None
+hp_space:
+  learning_parameters.learning_rate:
+    param_type: quniform
+    range:
+      - 0.001
+      - 0.1
+      - 0.001
+  learning_parameters.batch_size:
+    param_type: qloguniform
+    range:
+      - 2
+      - 4
+      - 2

external/mmdetection/configs/custom-counting-instance-seg/efficientnetb2b_maskrcnn/hpo_config.yaml

@@ -0,0 +1,160 @@
+_base_ = [
+    './coco_data_pipeline.py'
+]
+
+model = dict(
+    type='MaskRCNN',
+    backbone=dict(
+        type='efficientnet_b2b',
+        out_indices=(2, 3, 4, 5),
+        frozen_stages=-1,
+        pretrained=True,
+        activation_cfg=dict(type='torch_swish'),
+        norm_cfg=dict(type='BN', requires_grad=True)),
+    neck=dict(
+        type='FPN',
+        in_channels=[24, 48, 120, 352],
+        out_channels=80,
+        num_outs=5),
+    rpn_head=dict(
+        type='RPNHead',
+        in_channels=80,
+        feat_channels=80,
+        anchor_generator=dict(
+            type='AnchorGenerator',
+            scales=[8],
+            ratios=[0.5, 1.0, 2.0],
+            strides=[4, 8, 16, 32, 64]),
+        bbox_coder=dict(
+            type='DeltaXYWHBBoxCoder',
+            target_means=[0.0, 0.0, 0.0, 0.0],
+            target_stds=[1.0, 1.0, 1.0, 1.0]),
+        loss_cls=dict(
+            type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
+        loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
+    roi_head=dict(
+        type='StandardRoIHead',
+        bbox_roi_extractor=dict(
+            type='SingleRoIExtractor',
+            roi_layer=dict(type='RoIAlign', output_size=7, sampling_ratio=0),
+            out_channels=80,
+            featmap_strides=[4, 8, 16, 32]),
+        bbox_head=dict(
+            type='Shared2FCBBoxHead',
+            in_channels=80,
+            fc_out_channels=1024,
+            roi_feat_size=7,
+            num_classes=80,
+            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]),
+            reg_class_agnostic=False,
+            loss_cls=dict(
+                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),
+            loss_bbox=dict(type='L1Loss', loss_weight=1.0)),
+        mask_roi_extractor=dict(
+            type='SingleRoIExtractor',
+            roi_layer=dict(type='RoIAlign', output_size=14, sampling_ratio=0),
+            out_channels=80,
+            featmap_strides=[4, 8, 16, 32]),
+        mask_head=dict(
+            type='FCNMaskHead',
+            num_convs=4,
+            in_channels=80,
+            conv_out_channels=80,
+            num_classes=80,
+            loss_mask=dict(
+                type='CrossEntropyLoss', use_mask=True, loss_weight=1.0))),
+    train_cfg=dict(
+        rpn=dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.7,
+                neg_iou_thr=0.3,
+                min_pos_iou=0.3,
+                match_low_quality=True,
+                ignore_iof_thr=-1,
+                gpu_assign_thr=300),
+            sampler=dict(
+                type='RandomSampler',
+                num=256,
+                pos_fraction=0.5,
+                neg_pos_ub=-1,
+                add_gt_as_proposals=False),
+            allowed_border=-1,
+            pos_weight=-1,
+            debug=False),
+        rpn_proposal=dict(
+            nms_across_levels=False,
+            nms_pre=2000,
+            nms_post=1000,
+            max_num=1000,
+            nms_thr=0.8,
+            min_bbox_size=0),
+        rcnn=dict(
+            assigner=dict(
+                type='MaxIoUAssigner',
+                pos_iou_thr=0.5,
+                neg_iou_thr=0.5,
+                min_pos_iou=0.5,
+                match_low_quality=True,
+                ignore_iof_thr=-1,
+                gpu_assign_thr=300),
+            sampler=dict(
+                type='RandomSampler',
+                num=256,
+                pos_fraction=0.25,
+                neg_pos_ub=-1,
+                add_gt_as_proposals=True),
+            mask_size=28,
+            pos_weight=-1,
+            debug=False)),
+    test_cfg=dict(
+        rpn=dict(
+            nms_across_levels=False,
+            nms_pre=800,
+            nms_post=500,
+            max_num=500,
+            nms_thr=0.8,
+            min_bbox_size=0),
+        rcnn=dict(
+            score_thr=0.05,
+            nms=dict(type='nms', iou_threshold=0.7),
+            max_per_img=500,
+            mask_thr_binary=0.5)))
+
+cudnn_benchmark = True
+evaluation = dict(interval=1, metric='mAP', save_best='mAP', iou_thr=[0.5 , 0.55, 0.6 , 0.65, 0.7 , 0.75, 0.8 , 0.85, 0.9, 0.95])
+optimizer = dict(type='SGD', lr=0.015, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+lr_config = dict(
+    policy='ReduceLROnPlateau',
+    metric='mAP',
+    patience=5,
+    iteration_patience=300,
+    interval=1,
+    min_lr=0.000001,
+    warmup='linear',
+    warmup_iters=200,
+    warmup_ratio=0.3333333333333333
+)
+runner = dict(type='EpochRunnerWithCancel', max_epochs=300)
+checkpoint_config = dict(interval=5)
+log_config = dict(
+    interval=1,
+    hooks=[
+        dict(type='TextLoggerHook'),
+        # dict(type='TensorboardLoggerHook'),
+    ])
+dist_params = dict(backend='nccl')
+log_level = 'INFO'
+load_from = 'https://storage.openvinotoolkit.org/repositories/openvino_training_extensions/models/instance_segmentation/v2/efficientnet_b2b-mask_rcnn-576x576.pth'
+resume_from = None
+workflow = [('train', 1)]
+work_dir = 'output'
+custom_hooks = [
+    dict(type='EarlyStoppingHook', patience=10, metric='mAP',
+         interval=1, priority=75, iteration_patience=0),
+]
+fp16 = dict(loss_scale=512.)

external/mmdetection/configs/custom-counting-instance-seg/efficientnetb2b_maskrcnn/model.py

@@ -0,0 +1,52 @@
+# Description.
+model_template_id: Custom_Counting_Instance_Segmentation_MaskRCNN_EfficientNetB2B
+name: MaskRCNN-EfficientNetB2B
+task_type: INSTANCE_SEGMENTATION
+task_family: VISION
+instantiation: "CLASS"
+summary: Counting algorithm provides counting of objects and object instance masks. This model is based on MaskRCNN-EfficientNetB2B which is faster in training and inference but less accurate.
+application: ~
+
+# Algo backend.
+framework: OTEDetection v2.9.1
+
+# Task implementations.
+entrypoints:
+  base: detection_tasks.apis.detection.OTEDetectionTrainingTask
+  openvino: detection_tasks.apis.detection.OpenVINODetectionTask
+  nncf: detection_tasks.apis.detection.OTEDetectionNNCFTask
+
+# Capabilities.
+capabilities:
+  - compute_representations
+
+# Hyperparameters.
+hyper_parameters:
+  base_path: ../../../../mmdetection/detection_tasks/apis/detection/configuration.yaml
+  parameter_overrides:
+    learning_parameters:
+      batch_size:
+        default_value: 4
+      learning_rate:
+        default_value: 0.015
+      learning_rate_warmup_iters:
+        default_value: 200
+      num_iters:
+        default_value: 300
+    nncf_optimization:
+      enable_quantization:
+        default_value: true
+      enable_pruning:
+        default_value: false
+      maximal_accuracy_degradation:
+        default_value: 1.0
+
+# Training resources.
+max_nodes: 1
+training_targets:
+  - GPU
+  - CPU
+
+# Stats.
+gigaflops: 68.48
+size: 13.27