final version, with final config files

install.sh

@@ -4,19 +4,20 @@
 # Criando um ambiente conda específico para rodar o mmdetection
 # estou usando esta versão do python pois não testei em outras
 # (mas pode ser que funcione em outras)
-conda create --name od python=3.7.7 -y
+conda create --name od_porto python=3.7 -y
 
 # Ativando o ambiente recém criado (confira para ver se ativou 
 # corretamente olhando para a linha de comando, deve começar
 # a aparecer um (od) no lugar de (base) no prompt
-conda activate od
+conda activate od_porto
 
 # Algumas coisas instalam com o conda mas outras precisam do pip
-conda install pytorch torchvision -c pytorch
+conda install pytorch torchvision torchaudio cudatoolkit=11.0 -c pytorch
 # Se der erro na linha abaixo pode ter relação com a instalação do CUDA
 # e da placa gráfica na máquina 
-pip install mmcv-full  
+pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu110/torch1.7.0/index.html  
 cd mmdetection
 pip install -r requirements/build.txt
-pip install -v -e .  # or "python setup.py develop"
-pip install seaborn
+pip install -r requirements/optional.txt
+pip install -r requirements/runtime.txt
+pip install -r requirements/tests.txt

mmdetection/configs/_base_/datasets/coco_detection.py

 dataset_type = 'CocoDataset'
 data_root = 'data/coco/'
-img_norm_cfg = dict(
-    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
 train_pipeline = [
     dict(type='LoadImageFromFile'),
     dict(type='LoadAnnotations', with_bbox=True),
@@ -32,17 +31,17 @@ data = dict(
     workers_per_gpu=2,
     train=dict(
         type=dataset_type,
-        ann_file=data_root + 'annotations/instances_train2017.json',
-        img_prefix=data_root + 'train2017/',
+        ann_file=data_root + 'train/_annotations.coco.json',
+        img_prefix=data_root + 'train/',
         pipeline=train_pipeline),
     val=dict(
         type=dataset_type,
-        ann_file=data_root + 'annotations/instances_val2017.json',
-        img_prefix=data_root + 'val2017/',
+        ann_file=data_root + 'valid/_annotations.coco.json',
+        img_prefix=data_root + 'valid/',
         pipeline=test_pipeline),
     test=dict(
         type=dataset_type,
-        ann_file=data_root + 'annotations/instances_val2017.json',
-        img_prefix=data_root + 'val2017/',
+        ann_file=data_root + 'test/_annotations.coco.json',
+        img_prefix=data_root + 'test/',
         pipeline=test_pipeline))
 evaluation = dict(interval=1, metric='bbox')

mmdetection/configs/_base_/datasets/voc0712.py

+# Eu precisei mexer neste arquivo pois estou usando apenas o formato VOC2007
+# e não o VOC2012. Estava dando um erro antes, por isso eu copio este
+# arquivo para a estrutura do mmdetection antes de rodar 
+
 # dataset settings
 dataset_type = 'VOCDataset'
 data_root = 'data/VOCdevkit/'

mmdetection/configs/pascal_voc/faster_rcnn_r50_fpn_1x_coco_pupa.py

+# Neste arquivo são definidos os vários hiperparâmetros do modelo. 
+# Atenção com o número de classes que pode mudar de um conjunto de
+# dados para outro. Geralmente usa-se uma classe a mais por conta
+# do fundo (background)
+
+_base_ = [
+    '../_base_/models/faster_rcnn_r50_fpn.py', '../_base_/datasets/coco_detection.py',
+    '../_base_/default_runtime.py'
+]
+
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
+# TROQUE AQUI PELA QUANTIDADE DE CLASSES DO SEU PROBLEMA (UMA A MAIS PARA O FUNDO SE NECESSÁRIO)
+model = dict(roi_head=dict(bbox_head=dict(num_classes=2)))
+# optimizer
+optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
+#optimizer = dict(type='Adam', lr=0.001, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+# actual epoch = 3 * 3 = 9
+lr_config = dict(policy='step', step=[3])
+# runtime settings
+total_epochs = 1000  # actual epoch = 4 * 3 = 12
+
+# checkpoint saving configuration
+checkpoint_config = dict(interval=250)

mmdetection/configs/pascal_voc/faster_rcnn_r50_fpn_1x_voc0712.py

+# Neste arquivo são definidos os vários hiperparâmetros do modelo. 
+# Atenção com o número de classes que pode mudar de um conjunto de
+# dados para outro. Geralmente usa-se uma classe a mais por conta
+# do fundo (background)
+
 _base_ = [
     '../_base_/models/faster_rcnn_r50_fpn.py', '../_base_/datasets/voc0712.py',
     '../_base_/default_runtime.py'
 ]
 
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
 # TROQUE AQUI PELA QUANTIDADE DE CLASSES DO SEU PROBLEMA (UMA A MAIS PARA O FUNDO SE NECESSÁRIO)
 model = dict(roi_head=dict(bbox_head=dict(num_classes=2)))
 # optimizer
-#optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
-optimizer = dict(type='Adam', lr=0.001, weight_decay=0.0001)
+optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
+#optimizer = dict(type='Adam', lr=0.001, weight_decay=0.0001)
 optimizer_config = dict(grad_clip=None)
 # learning policy
 # actual epoch = 3 * 3 = 9
 lr_config = dict(policy='step', step=[3])
 # runtime settings
 total_epochs = 1000  # actual epoch = 4 * 3 = 12
+
+# checkpoint saving configuration
+checkpoint_config = dict(interval=250)

mmdetection/configs/pascal_voc/mask_rcnn_r50_fpn_1x_coco_pupa.py

+_base_ = [
+    '../_base_/models/mask_rcnn_r50_fpn.py',
+    '../_base_/datasets/coco_instance.py',
+    '../_base_/default_runtime.py'
+]
+
+train_pipeline = [
+    dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
+]
+
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
+#optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
+optimizer = dict(type='Adam', lr=0.001, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+# actual epoch = 3 * 3 = 9
+lr_config = dict(policy='step', step=[3])
+# runtime settings
+total_epochs = 1000  # actual epoch = 4 * 3 = 12
+
+# checkpoint saving configuration
+checkpoint_config = dict(interval=250)

mmdetection/configs/pascal_voc/mask_rcnn_r50_fpn_1x_voc0712.py

+_base_ = [
+    '../_base_/models/mask_rcnn_r50_fpn.py',
+    '../_base_/datasets/voc0712.py',
+    '../_base_/default_runtime.py'
+]
+
+train_pipeline = [
+    dict(type='LoadAnnotations', with_bbox=True, with_mask=True),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels', 'gt_masks']),
+]
+
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
+#optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
+optimizer = dict(type='Adam', lr=0.001, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+# actual epoch = 3 * 3 = 9
+lr_config = dict(policy='step', step=[3])
+# runtime settings
+total_epochs = 1000  # actual epoch = 4 * 3 = 12
+
+# checkpoint saving configuration
+checkpoint_config = dict(interval=250)

mmdetection/configs/pascal_voc/retinanet_r50_fpn_1x_coco_pupa.py

+# Neste arquivo são definidos os vários hiperparâmetros do modelo. 
+# Atenção com o número de classes que pode mudar de um conjunto de
+# dados para outro. Geralmente usa-se uma classe a mais por conta
+# do fundo (background)
+
+_base_ = [
+    '../_base_/models/retinanet_r50_fpn.py', '../_base_/datasets/coco_detection.py',
+    '../_base_/default_runtime.py'
+]
+
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
+# TROQUE AQUI PELA QUANTIDADE DE CLASSES DO SEU PROBLEMA (UMA A MAIS PARA O FUNDO SE NECESSÁRIO)
+model = dict(bbox_head=dict(num_classes=2))
+
+data = dict(
+    samples_per_gpu=2,
+    workers_per_gpu=2)
+# optimizer
+optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict(grad_clip=None)
+# learning policy
+# actual epoch = 3 * 3 = 9
+lr_config = dict(policy='step', step=[3])
+# runtime settings
+total_epochs = 1000
+
+# checkpoint saving configuration
+checkpoint_config = dict(interval=250)
+# mexer mais aqui depois pra ver o pq q ta dando esses erros de nao salvar resutlado
+# 2021-03-12 16:31:10,797 - mmdet - ERROR - The testing results of the whole dataset is empty.
+# acontece depois da primiera epoca
+

mmdetection/configs/pascal_voc/retinanet_r50_fpn_1x_voc0712.py

+# Neste arquivo são definidos os vários hiperparâmetros do modelo. 
+# Atenção com o número de classes que pode mudar de um conjunto de
+# dados para outro. Geralmente usa-se uma classe a mais por conta
+# do fundo (background)
+
 _base_ = [
     '../_base_/models/retinanet_r50_fpn.py', '../_base_/datasets/voc0712.py',
     '../_base_/default_runtime.py'
 ]
+
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
+# TROQUE AQUI PELA QUANTIDADE DE CLASSES DO SEU PROBLEMA (UMA A MAIS PARA O FUNDO SE NECESSÁRIO)
 model = dict(bbox_head=dict(num_classes=2))
 # optimizer
 optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
@@ -10,4 +18,6 @@ optimizer_config = dict(grad_clip=None)
 # actual epoch = 3 * 3 = 9
 lr_config = dict(policy='step', step=[3])
 # runtime settings
-total_epochs = 40  # actual epoch = 4 * 3 = 12
+total_epochs = 1000
+# checkpoint saving configuration
+checkpoint_config = dict(interval=250)

mmdetection/configs/pascal_voc/ssd300_coco_pupa.py

+_base_ = [
+    '../_base_/models/ssd300.py', '../_base_/datasets/coco_detection.py',
+    '../_base_/default_runtime.py'
+]
+model = dict(
+    bbox_head=dict(
+        num_classes=2, anchor_generator=dict(basesize_ratio_range=(0.2,
+                                                                    0.9))))
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile', to_float32=True),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(
+        type='PhotoMetricDistortion',
+        brightness_delta=32,
+        contrast_range=(0.5, 1.5),
+        saturation_range=(0.5, 1.5),
+        hue_delta=18),
+    dict(
+        type='Expand',
+        mean=img_norm_cfg['mean'],
+        to_rgb=img_norm_cfg['to_rgb'],
+        ratio_range=(1, 4)),
+    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=(300, 300), keep_ratio=False),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='RandomFlip', flip_ratio=0.5),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']),
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(300, 300),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=False),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img']),
+        ])
+]
+data = dict(
+    samples_per_gpu=8,
+    workers_per_gpu=3,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'train/_annotations.coco.json',
+        img_prefix=data_root + 'train/',
+        pipeline=train_pipeline),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'valid/_annotations.coco.json',
+        img_prefix=data_root + 'valid/',
+        pipeline=test_pipeline),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'test/_annotations.coco.json',
+        img_prefix=data_root + 'test/',
+        pipeline=test_pipeline))
+# optimizer
+optimizer = dict(type='SGD', lr=1e-3, momentum=0.9, weight_decay=5e-4)
+optimizer_config = dict()
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=0.001,
+    step=[16, 20])
+
+# runtime settings
+total_epochs = 1000
+
+
+# checkpoint saving configuration
+checkpoint_config = dict(interval=250)

mmdetection/configs/pascal_voc/ssd300_voc0712.py

-_base_ = [
-    '../_base_/models/ssd300.py', '../_base_/datasets/voc0712.py',
-    '../_base_/default_runtime.py'
-]
-model = dict(
-    bbox_head=dict(
-        num_classes=20, anchor_generator=dict(basesize_ratio_range=(0.2,
-                                                                    0.9))))
-# dataset settings
-dataset_type = 'VOCDataset'
-data_root = 'data/VOCdevkit/'
-img_norm_cfg = dict(mean=[123.675, 116.28, 103.53], std=[1, 1, 1], to_rgb=True)
-train_pipeline = [
-    dict(type='LoadImageFromFile', to_float32=True),
-    dict(type='LoadAnnotations', with_bbox=True),
-    dict(
-        type='PhotoMetricDistortion',
-        brightness_delta=32,
-        contrast_range=(0.5, 1.5),
-        saturation_range=(0.5, 1.5),
-        hue_delta=18),
-    dict(
-        type='Expand',
-        mean=img_norm_cfg['mean'],
-        to_rgb=img_norm_cfg['to_rgb'],
-        ratio_range=(1, 4)),
-    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=(300, 300), keep_ratio=False),
-    dict(type='Normalize', **img_norm_cfg),
-    dict(type='RandomFlip', flip_ratio=0.5),
-    dict(type='DefaultFormatBundle'),
-    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']),
-]
-test_pipeline = [
-    dict(type='LoadImageFromFile'),
-    dict(
-        type='MultiScaleFlipAug',
-        img_scale=(300, 300),
-        flip=False,
-        transforms=[
-            dict(type='Resize', keep_ratio=False),
-            dict(type='Normalize', **img_norm_cfg),
-            dict(type='ImageToTensor', keys=['img']),
-            dict(type='Collect', keys=['img']),
-        ])
-]
-data = dict(
-    samples_per_gpu=8,
-    workers_per_gpu=3,
-    train=dict(
-        type='RepeatDataset', times=10, dataset=dict(pipeline=train_pipeline)),
-    val=dict(pipeline=test_pipeline),
-    test=dict(pipeline=test_pipeline))
-# optimizer
-optimizer = dict(type='SGD', lr=1e-3, momentum=0.9, weight_decay=5e-4)
-optimizer_config = dict()
-# learning policy
-lr_config = dict(
-    policy='step',
-    warmup='linear',
-    warmup_iters=500,
-    warmup_ratio=0.001,
-    step=[16, 20])
-checkpoint_config = dict(interval=1)
-# runtime settings
-total_epochs = 24
+_base_ = [
+    '../_base_/models/ssd300.py', '../_base_/datasets/voc0712.py',
+    '../_base_/default_runtime.py'
+]
+model = dict(
+    bbox_head=dict(
+        num_classes=2, anchor_generator=dict(basesize_ratio_range=(0.2,
+                                                                    0.9))))
+# dataset settings
+dataset_type = 'VOCDataset'
+data_root = 'data/VOCdevkit/'
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile', to_float32=True),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(
+        type='PhotoMetricDistortion',
+        brightness_delta=32,
+        contrast_range=(0.5, 1.5),
+        saturation_range=(0.5, 1.5),
+        hue_delta=18),
+    dict(
+        type='Expand',
+        mean=img_norm_cfg['mean'],
+        to_rgb=img_norm_cfg['to_rgb'],
+        ratio_range=(1, 4)),
+    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=(300, 300), keep_ratio=False),
+    dict(type='Normalize', **img_norm_cfg),
+    dict(type='RandomFlip', flip_ratio=0.5),
+    dict(type='DefaultFormatBundle'),
+    dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']),
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(300, 300),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=False),
+            dict(type='Normalize', **img_norm_cfg),
+            dict(type='ImageToTensor', keys=['img']),
+            dict(type='Collect', keys=['img']),
+        ])
+]
+data = dict(
+    train=dict(
+    dataset=dict(pipeline=train_pipeline)),
+    val=dict(pipeline=test_pipeline),
+    test=dict(pipeline=test_pipeline))
+# optimizer
+optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001)
+optimizer_config = dict()
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=500,
+    warmup_ratio=0.001,
+    step=[16, 20])
+
+# runtime settings
+total_epochs = 1000
+
+
+# checkpoint saving configuration
+checkpoint_config = dict(interval=250)

mmdetection/configs/pascal_voc/yolov3_d53_mstrain-608_273e_coco_pupa.py

+_base_ = '../_base_/default_runtime.py'
+# model settings
+model = dict(
+    type='YOLOV3',
+    pretrained='open-mmlab://darknet53',
+    backbone=dict(type='Darknet', depth=53, out_indices=(3, 4, 5)),
+    neck=dict(
+        type='YOLOV3Neck',
+        num_scales=3,
+        in_channels=[1024, 512, 256],
+        out_channels=[512, 256, 128]),
+    bbox_head=dict(
+        type='YOLOV3Head',
+        num_classes=2,
+        in_channels=[512, 256, 128],
+        out_channels=[1024, 512, 256],
+        anchor_generator=dict(
+            type='YOLOAnchorGenerator',
+            base_sizes=[[(116, 90), (156, 198), (373, 326)],
+                        [(30, 61), (62, 45), (59, 119)],
+                        [(10, 13), (16, 30), (33, 23)]],
+            strides=[32, 16, 8]),
+        bbox_coder=dict(type='YOLOBBoxCoder'),
+        featmap_strides=[32, 16, 8],
+        loss_cls=dict(
+            type='CrossEntropyLoss',
+            use_sigmoid=True,
+            loss_weight=1.0,
+            reduction='sum'),
+        loss_conf=dict(
+            type='CrossEntropyLoss',
+            use_sigmoid=True,
+            loss_weight=1.0,
+            reduction='sum'),
+        loss_xy=dict(
+            type='CrossEntropyLoss',
+            use_sigmoid=True,
+            loss_weight=2.0,
+            reduction='sum'),
+        loss_wh=dict(type='MSELoss', loss_weight=2.0, reduction='sum')))
+# training and testing settings
+train_cfg = dict(
+    assigner=dict(
+        type='GridAssigner', pos_iou_thr=0.5, neg_iou_thr=0.5, min_pos_iou=0))
+test_cfg = dict(
+    nms_pre=1000,
+    min_bbox_size=0,
+    score_thr=0.05,
+    conf_thr=0.005,
+    nms=dict(type='nms', iou_threshold=0.45),
+    max_per_img=100)
+# dataset settings
+dataset_type = 'CocoDataset'
+data_root = 'data/coco/'
+img_norm_cfg = dict(mean=[154.04, 159.83, 181.41], std=[13.59, 17.70, 25.08], to_rgb=True)
+train_pipeline = [
+    dict(type='LoadImageFromFile', to_float32=True),
+    dict(type='LoadAnnotations', with_bbox=True),
+    dict(type='PhotoMetricDistortion'),
+    dict(
+        type='Expand',
+        mean=img_norm_cfg['mean'],
+        to_rgb=img_norm_cfg['to_rgb'],
+        ratio_range=(1, 2)),
+    dict(
+        type='MinIoURandomCrop',
+        min_ious=(0.4, 0.5, 0.6, 0.7, 0.8, 0.9),
+        min_crop_size=0.3),
+    dict(type='Resize', img_scale=[(320, 320), (608, 608)], keep_ratio=True),
+    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'])
+]
+test_pipeline = [
+    dict(type='LoadImageFromFile'),
+    dict(
+        type='MultiScaleFlipAug',
+        img_scale=(608, 608),
+        flip=False,
+        transforms=[
+            dict(type='Resize', keep_ratio=True),
+            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=8,
+    workers_per_gpu=3,
+    train=dict(
+        type=dataset_type,
+        ann_file=data_root + 'train/_annotations.coco.json',
+        img_prefix=data_root + 'train/',
+        pipeline=train_pipeline),
+    val=dict(
+        type=dataset_type,
+        ann_file=data_root + 'valid/_annotations.coco.json',
+        img_prefix=data_root + 'valid/',
+        pipeline=test_pipeline),
+    test=dict(
+        type=dataset_type,
+        ann_file=data_root + 'test/_annotations.coco.json',
+        img_prefix=data_root + 'test/',
+        pipeline=test_pipeline))
+
+# optimizer
+optimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0005)
+optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))
+# learning policy
+lr_config = dict(
+    policy='step',
+    warmup='linear',
+    warmup_iters=2000,  # same as burn-in in darknet