[Feature] Add nanodet-plus and support grad clip. (#362)

* [Feature] Add nanodet-plus and support grad clip. * fix forward * add cfg * refactor * unit test
RangiLyu · Dec 23, 2021 · 5967031 · 5967031
1 parent 2ca268a
commit 5967031
Show file tree

Hide file tree

Showing 18 changed files with 772 additions and 23 deletions.
diff --git a/config/nanodet-plus-m-1.5x_320.yml b/config/nanodet-plus-m-1.5x_320.yml
@@ -0,0 +1,131 @@
+# nanodet-plus-m-1.5x_320
+# COCO mAP(0.5:0.95) = 0.299
+# AP_50 = 0.454
+# AP_75 = 0.312
+# AP_small = 0.102
+# AP_m = 0.309
+# AP_l = 0.493
+save_dir: workspace/nanodet-plus-m-1.5x_320
+model:
+ weight_averager:
+ name: ExpMovingAverager
+ decay: 0.9998
+ arch:
+ name: NanoDetPlus
+ detach_epoch: 10
+ backbone:
+ name: ShuffleNetV2
+ model_size: 1.5x
+ out_stages: [2,3,4]
+ activation: LeakyReLU
+ fpn:
+ name: GhostPAN
+ in_channels: [176, 352, 704]
+ out_channels: 128
+ kernel_size: 5
+ num_extra_level: 1
+ use_depthwise: True
+ activation: LeakyReLU
+ head:
+ name: NanoDetPlusHead
+ num_classes: 80
+ input_channel: 128
+ feat_channels: 128
+ stacked_convs: 2
+ kernel_size: 5
+ strides: [8, 16, 32, 64]
+ activation: LeakyReLU
+ reg_max: 7
+ norm_cfg:
+ type: BN
+ loss:
+ loss_qfl:
+ name: QualityFocalLoss
+ use_sigmoid: True
+ beta: 2.0
+ loss_weight: 1.0
+ loss_dfl:
+ name: DistributionFocalLoss
+ loss_weight: 0.25
+ loss_bbox:
+ name: GIoULoss
+ loss_weight: 2.0
+ # Auxiliary head, only use in training time.
+ aux_head:
+ name: SimpleConvHead
+ num_classes: 80
+ input_channel: 256
+ feat_channels: 256
+ stacked_convs: 4
+ strides: [8, 16, 32, 64]
+ activation: LeakyReLU
+ reg_max: 7
+data:
+ train:
+ name: CocoDataset
+ img_path: coco/train2017
+ ann_path: coco/annotations/instances_train2017.json
+ input_size: [320,320] #[w,h]
+ keep_ratio: False
+ pipeline:
+ perspective: 0.0
+ scale: [0.6, 1.4]
+ stretch: [[0.8, 1.2], [0.8, 1.2]]
+ rotation: 0
+ shear: 0
+ translate: 0.2
+ flip: 0.5
+ brightness: 0.2
+ contrast: [0.6, 1.4]
+ saturation: [0.5, 1.2]
+ normalize: [[103.53, 116.28, 123.675], [57.375, 57.12, 58.395]]
+ val:
+ name: CocoDataset
+ img_path: coco/val2017
+ ann_path: coco/annotations/instances_val2017.json
+ input_size: [320,320] #[w,h]
+ keep_ratio: False
+ pipeline:
+ normalize: [[103.53, 116.28, 123.675], [57.375, 57.12, 58.395]]
+device:
+ gpu_ids: [0]
+ workers_per_gpu: 10
+ batchsize_per_gpu: 96
+schedule:
+# resume:
+# load_model:
+ optimizer:
+ name: AdamW
+ lr: 0.001
+ weight_decay: 0.05
+ warmup:
+ name: linear
+ steps: 500
+ ratio: 0.0001
+ total_epochs: 300
+ lr_schedule:
+ name: CosineAnnealingLR
+ T_max: 300
+ eta_min: 0.00005
+ val_intervals: 10
+grad_clip: 35
+evaluator:
+ name: CocoDetectionEvaluator
+ save_key: mAP
+log:
+ interval: 50
+
+class_names: ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
+ 'train', 'truck', 'boat', 'traffic_light', 'fire_hydrant',
+ 'stop_sign', 'parking_meter', 'bench', 'bird', 'cat', 'dog',
+ 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe',
+ 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee',
+ 'skis', 'snowboard', 'sports_ball', 'kite', 'baseball_bat',
+ 'baseball_glove', 'skateboard', 'surfboard', 'tennis_racket',
+ 'bottle', 'wine_glass', 'cup', 'fork', 'knife', 'spoon', 'bowl',
+ 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot',
+ 'hot_dog', 'pizza', 'donut', 'cake', 'chair', 'couch',
+ 'potted_plant', 'bed', 'dining_table', 'toilet', 'tv', 'laptop',
+ 'mouse', 'remote', 'keyboard', 'cell_phone', 'microwave',
+ 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock',
+ 'vase', 'scissors', 'teddy_bear', 'hair_drier', 'toothbrush']
diff --git a/config/nanodet-plus-m-1.5x_416.yml b/config/nanodet-plus-m-1.5x_416.yml
@@ -0,0 +1,131 @@
+# nanodet-plus-m-1.5x_416
+# COCO mAP(0.5:0.95) = 0.341
+# AP_50 = 0.506
+# AP_75 = 0.357
+# AP_small = 0.143
+# AP_m = 0.363
+# AP_l = 0.539
+save_dir: workspace/nanodet-plus-m-1.5x_416
+model:
+ weight_averager:
+ name: ExpMovingAverager
+ decay: 0.9998
+ arch:
+ name: NanoDetPlus
+ detach_epoch: 10
+ backbone:
+ name: ShuffleNetV2
+ model_size: 1.5x
+ out_stages: [2,3,4]
+ activation: LeakyReLU
+ fpn:
+ name: GhostPAN
+ in_channels: [176, 352, 704]
+ out_channels: 128
+ kernel_size: 5
+ num_extra_level: 1
+ use_depthwise: True
+ activation: LeakyReLU
+ head:
+ name: NanoDetPlusHead
+ num_classes: 80
+ input_channel: 128
+ feat_channels: 128
+ stacked_convs: 2
+ kernel_size: 5
+ strides: [8, 16, 32, 64]
+ activation: LeakyReLU
+ reg_max: 7
+ norm_cfg:
+ type: BN
+ loss:
+ loss_qfl:
+ name: QualityFocalLoss
+ use_sigmoid: True
+ beta: 2.0
+ loss_weight: 1.0
+ loss_dfl:
+ name: DistributionFocalLoss
+ loss_weight: 0.25
+ loss_bbox:
+ name: GIoULoss
+ loss_weight: 2.0
+ # Auxiliary head, only use in training time.
+ aux_head:
+ name: SimpleConvHead
+ num_classes: 80
+ input_channel: 256
+ feat_channels: 256
+ stacked_convs: 4
+ strides: [8, 16, 32, 64]
+ activation: LeakyReLU
+ reg_max: 7
+data:
+ train:
+ name: CocoDataset
+ img_path: coco/train2017
+ ann_path: coco/annotations/instances_train2017.json
+ input_size: [416,416] #[w,h]
+ keep_ratio: False
+ pipeline:
+ perspective: 0.0
+ scale: [0.6, 1.4]
+ stretch: [[0.8, 1.2], [0.8, 1.2]]
+ rotation: 0
+ shear: 0
+ translate: 0.2
+ flip: 0.5
+ brightness: 0.2
+ contrast: [0.6, 1.4]
+ saturation: [0.5, 1.2]
+ normalize: [[103.53, 116.28, 123.675], [57.375, 57.12, 58.395]]
+ val:
+ name: CocoDataset
+ img_path: coco/val2017
+ ann_path: coco/annotations/instances_val2017.json
+ input_size: [416,416] #[w,h]
+ keep_ratio: False
+ pipeline:
+ normalize: [[103.53, 116.28, 123.675], [57.375, 57.12, 58.395]]
+device:
+ gpu_ids: [0]
+ workers_per_gpu: 10
+ batchsize_per_gpu: 96
+schedule:
+# resume:
+# load_model:
+ optimizer:
+ name: AdamW
+ lr: 0.001
+ weight_decay: 0.05
+ warmup:
+ name: linear
+ steps: 500
+ ratio: 0.0001
+ total_epochs: 300
+ lr_schedule:
+ name: CosineAnnealingLR
+ T_max: 300
+ eta_min: 0.00005
+ val_intervals: 10
+grad_clip: 35
+evaluator:
+ name: CocoDetectionEvaluator
+ save_key: mAP
+log:
+ interval: 50
+
+class_names: ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
+ 'train', 'truck', 'boat', 'traffic_light', 'fire_hydrant',
+ 'stop_sign', 'parking_meter', 'bench', 'bird', 'cat', 'dog',
+ 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe',
+ 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee',
+ 'skis', 'snowboard', 'sports_ball', 'kite', 'baseball_bat',
+ 'baseball_glove', 'skateboard', 'surfboard', 'tennis_racket',
+ 'bottle', 'wine_glass', 'cup', 'fork', 'knife', 'spoon', 'bowl',
+ 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot',
+ 'hot_dog', 'pizza', 'donut', 'cake', 'chair', 'couch',
+ 'potted_plant', 'bed', 'dining_table', 'toilet', 'tv', 'laptop',
+ 'mouse', 'remote', 'keyboard', 'cell_phone', 'microwave',
+ 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock',
+ 'vase', 'scissors', 'teddy_bear', 'hair_drier', 'toothbrush']
diff --git a/config/nanodet-plus-m_320.yml b/config/nanodet-plus-m_320.yml
@@ -0,0 +1,131 @@
+# nanodet-plus-m_320
+# COCO mAP(0.5:0.95) = 0.270
+# AP_50 = 0.418
+# AP_75 = 0.281
+# AP_small = 0.083
+# AP_m = 0.278
+# AP_l = 0.451
+save_dir: workspace/nanodet-plus-m_320
+model:
+ weight_averager:
+ name: ExpMovingAverager
+ decay: 0.9998
+ arch:
+ name: NanoDetPlus
+ detach_epoch: 10
+ backbone:
+ name: ShuffleNetV2
+ model_size: 1.0x
+ out_stages: [2,3,4]
+ activation: LeakyReLU
+ fpn:
+ name: GhostPAN
+ in_channels: [116, 232, 464]
+ out_channels: 96
+ kernel_size: 5
+ num_extra_level: 1
+ use_depthwise: True
+ activation: LeakyReLU
+ head:
+ name: NanoDetPlusHead
+ num_classes: 80
+ input_channel: 96
+ feat_channels: 96
+ stacked_convs: 2
+ kernel_size: 5
+ strides: [8, 16, 32, 64]
+ activation: LeakyReLU
+ reg_max: 7
+ norm_cfg:
+ type: BN
+ loss:
+ loss_qfl:
+ name: QualityFocalLoss
+ use_sigmoid: True
+ beta: 2.0
+ loss_weight: 1.0
+ loss_dfl:
+ name: DistributionFocalLoss
+ loss_weight: 0.25
+ loss_bbox:
+ name: GIoULoss
+ loss_weight: 2.0
+ # Auxiliary head, only use in training time.
+ aux_head:
+ name: SimpleConvHead
+ num_classes: 80
+ input_channel: 192
+ feat_channels: 192
+ stacked_convs: 4
+ strides: [8, 16, 32, 64]
+ activation: LeakyReLU
+ reg_max: 7
+data:
+ train:
+ name: CocoDataset
+ img_path: coco/train2017
+ ann_path: coco/annotations/instances_train2017.json
+ input_size: [320,320] #[w,h]
+ keep_ratio: False
+ pipeline:
+ perspective: 0.0
+ scale: [0.6, 1.4]
+ stretch: [[0.8, 1.2], [0.8, 1.2]]
+ rotation: 0
+ shear: 0
+ translate: 0.2
+ flip: 0.5
+ brightness: 0.2
+ contrast: [0.6, 1.4]
+ saturation: [0.5, 1.2]
+ normalize: [[103.53, 116.28, 123.675], [57.375, 57.12, 58.395]]
+ val:
+ name: CocoDataset
+ img_path: coco/val2017
+ ann_path: coco/annotations/instances_val2017.json
+ input_size: [320,320] #[w,h]
+ keep_ratio: False
+ pipeline:
+ normalize: [[103.53, 116.28, 123.675], [57.375, 57.12, 58.395]]
+device:
+ gpu_ids: [0]
+ workers_per_gpu: 10
+ batchsize_per_gpu: 96
+schedule:
+# resume:
+# load_model:
+ optimizer:
+ name: AdamW
+ lr: 0.001
+ weight_decay: 0.05
+ warmup:
+ name: linear
+ steps: 500
+ ratio: 0.0001
+ total_epochs: 300
+ lr_schedule:
+ name: CosineAnnealingLR
+ T_max: 300
+ eta_min: 0.00005
+ val_intervals: 10
+grad_clip: 35
+evaluator:
+ name: CocoDetectionEvaluator
+ save_key: mAP
+log:
+ interval: 50
+
+class_names: ['person', 'bicycle', 'car', 'motorcycle', 'airplane', 'bus',
+ 'train', 'truck', 'boat', 'traffic_light', 'fire_hydrant',
+ 'stop_sign', 'parking_meter', 'bench', 'bird', 'cat', 'dog',
+ 'horse', 'sheep', 'cow', 'elephant', 'bear', 'zebra', 'giraffe',
+ 'backpack', 'umbrella', 'handbag', 'tie', 'suitcase', 'frisbee',
+ 'skis', 'snowboard', 'sports_ball', 'kite', 'baseball_bat',
+ 'baseball_glove', 'skateboard', 'surfboard', 'tennis_racket',
+ 'bottle', 'wine_glass', 'cup', 'fork', 'knife', 'spoon', 'bowl',
+ 'banana', 'apple', 'sandwich', 'orange', 'broccoli', 'carrot',
+ 'hot_dog', 'pizza', 'donut', 'cake', 'chair', 'couch',
+ 'potted_plant', 'bed', 'dining_table', 'toilet', 'tv', 'laptop',
+ 'mouse', 'remote', 'keyboard', 'cell_phone', 'microwave',
+ 'oven', 'toaster', 'sink', 'refrigerator', 'book', 'clock',
+ 'vase', 'scissors', 'teddy_bear', 'hair_drier', 'toothbrush']