一些目標(biāo)檢測(cè)技巧

train_pipeline = [    dict(type='LoadImageFromFile'),    dict(type='LoadAnnotations', with_bbox=True),    dict(type='Resize', img_scale=(1333, 800), 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=(1333, 800),        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']),        ])]

def pre_pipeline(self, results):        results['img_prefix'] = self.img_prefix        results['seg_prefix'] = self.seg_prefix        results['proposal_file'] = self.proposal_file        results['bbox_fields'] = []        results['mask_fields'] = []

train_pipeline = [    dict(type='LoadImageFromFile'),    dict(type='LoadAnnotations', with_bbox=True),    dict(type='Resize', img_scale=(1333, 800), keep_ratio=True), #這里可以更換多尺度[(),()]    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=(1333, 800),        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']),        ])]

rcnn=[        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.4, # 更換                neg_iou_thr=0.4,                min_pos_iou=0.4,                ignore_iof_thr=-1),            sampler=dict(                type='OHEMSampler',                num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False),        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.5,                neg_iou_thr=0.5,                min_pos_iou=0.5,                ignore_iof_thr=-1),            sampler=dict(                type='OHEMSampler', # 解決難易樣本，也解決了正負(fù)樣本比例問(wèn)題。num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False),        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.6,                neg_iou_thr=0.6,                min_pos_iou=0.6,                ignore_iof_thr=-1),            sampler=dict(                type='OHEMSampler',                num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False)    ],    stage_loss_weights=[1, 0.5, 0.25])

test_cfg = dict(    rpn=dict(        nms_across_levels=False,        nms_pre=1000,        nms_post=1000,        max_num=1000,        nms_thr=0.7,        min_bbox_size=0),    rcnn=dict(        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=20)) # 這里可以換為sof_tnms

# model settingsmodel = dict(  type='FasterRCNN',                         # model類(lèi)型    pretrained='modelzoo://resnet50',          # 預(yù)訓(xùn)練模型：imagenet-resnet50    backbone=dict(        type='ResNet',                         # backbone類(lèi)型        depth=50,                              # 網(wǎng)絡(luò)層數(shù)        num_stages=4,                          # resnet的stage數(shù)量        out_indices=(0, 1, 2, 3),              # 輸出的stage的序號(hào)        frozen_stages=1,                       # 凍結(jié)的stage數(shù)量，即該stage不更新參數(shù)，-1表示所有的stage都更新參數(shù)        style='pytorch'),                      # 網(wǎng)絡(luò)風(fēng)格：如果設(shè)置pytorch，則stride為2的層是conv3x3的卷積層；如果設(shè)置caffe，則stride為2的層是第一個(gè)conv1x1的卷積層    neck=dict(        type='FPN',                            # neck類(lèi)型        in_channels=[256, 512, 1024, 2048],    # 輸入的各個(gè)stage的通道數(shù)        out_channels=256,                      # 輸出的特征層的通道數(shù)        num_outs=5),                           # 輸出的特征層的數(shù)量    rpn_head=dict(        type='RPNHead',                        # RPN網(wǎng)絡(luò)類(lèi)型        in_channels=256,                       # RPN網(wǎng)絡(luò)的輸入通道數(shù)        feat_channels=256,                     # 特征層的通道數(shù)        anchor_scales=[8],                     # 生成的anchor的baselen，baselen = sqrt(w*h)，w和h為anchor的寬和高        anchor_ratios=[0.5, 1.0, 2.0],         # anchor的寬高比        anchor_strides=[4, 8, 16, 32, 64],     # 在每個(gè)特征層上的anchor的步長(zhǎng)（對(duì)應(yīng)于原圖）        target_means=[.0, .0, .0, .0],         # 均值        target_stds=[1.0, 1.0, 1.0, 1.0],      # 方差        use_sigmoid_cls=True),                 # 是否使用sigmoid來(lái)進(jìn)行分類(lèi)，如果False則使用softmax來(lái)分類(lèi)    bbox_roi_extractor=dict(        type='SingleRoIExtractor',                                   # RoIExtractor類(lèi)型        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),   # ROI具體參數(shù)：ROI類(lèi)型為ROIalign，輸出尺寸為7，sample數(shù)為2        out_channels=256,                                            # 輸出通道數(shù)        featmap_strides=[4, 8, 16, 32]),                             # 特征圖的步長(zhǎng)    bbox_head=dict(        type='SharedFCBBoxHead',                     # 全連接層類(lèi)型        num_fcs=2,                                   # 全連接層數(shù)量        in_channels=256,                             # 輸入通道數(shù)        fc_out_channels=1024,                        # 輸出通道數(shù)        roi_feat_size=7,                             # ROI特征層尺寸        num_classes=81,                              # 分類(lèi)器的類(lèi)別數(shù)量+1，+1是因?yàn)槎嗔艘粋€(gè)背景的類(lèi)別        target_means=[0., 0., 0., 0.],               # 均值        target_stds=[0.1, 0.1, 0.2, 0.2],            # 方差        reg_class_agnostic=False))                   # 是否采用class_agnostic的方式來(lái)預(yù)測(cè)，class_agnostic表示輸出bbox時(shí)只考慮其是否為前景，后續(xù)分類(lèi)的時(shí)候再根據(jù)該bbox在網(wǎng)絡(luò)中的類(lèi)別得分來(lái)分類(lèi)，也就是說(shuō)一個(gè)框可以對(duì)應(yīng)多個(gè)類(lèi)別# model training and testing settingstrain_cfg = dict(    rpn=dict(        assigner=dict(            type='MaxIoUAssigner',            # RPN網(wǎng)絡(luò)的正負(fù)樣本劃分            pos_iou_thr=0.7,                  # 正樣本的iou閾值            neg_iou_thr=0.3,                  # 負(fù)樣本的iou閾值            min_pos_iou=0.3,                  # 正樣本的iou最小值。如果assign給ground truth的anchors中最大的IOU低于0.3，則忽略所有的anchors，否則保留最大IOU的anchor            ignore_iof_thr=-1),               # 忽略bbox的閾值，當(dāng)ground truth中包含需要忽略的bbox時(shí)使用，-1表示不忽略        sampler=dict(            type='RandomSampler',             # 正負(fù)樣本提取器類(lèi)型            num=256,                          # 需提取的正負(fù)樣本數(shù)量            pos_fraction=0.5,                 # 正樣本比例            neg_pos_ub=-1,                    # 最大負(fù)樣本比例，大于該比例的負(fù)樣本忽略，-1表示不忽略            add_gt_as_proposals=False),       # 把ground truth加入proposal作為正樣本        allowed_border=0,                     # 允許在bbox周?chē)鈹U(kuò)一定的像素        pos_weight=-1,                        # 正樣本權(quán)重，-1表示不改變?cè)嫉臋?quán)重        smoothl1_beta=1 / 9.0,                # 平滑L1系數(shù)        debug=False),                         # debug模式    rcnn=dict(        assigner=dict(            type='MaxIoUAssigner',            # RCNN網(wǎng)絡(luò)正負(fù)樣本劃分            pos_iou_thr=0.5,                  # 正樣本的iou閾值            neg_iou_thr=0.5,                  # 負(fù)樣本的iou閾值            min_pos_iou=0.5,                  # 正樣本的iou最小值。如果assign給ground truth的anchors中最大的IOU低于0.3，則忽略所有的anchors，否則保留最大IOU的anchor            ignore_iof_thr=-1),               # 忽略bbox的閾值，當(dāng)ground truth中包含需要忽略的bbox時(shí)使用，-1表示不忽略        sampler=dict(            type='RandomSampler',             # 正負(fù)樣本提取器類(lèi)型            num=512,                          # 需提取的正負(fù)樣本數(shù)量            pos_fraction=0.25,                # 正樣本比例            neg_pos_ub=-1,                    # 最大負(fù)樣本比例，大于該比例的負(fù)樣本忽略，-1表示不忽略            add_gt_as_proposals=True),        # 把ground truth加入proposal作為正樣本        pos_weight=-1,                        # 正樣本權(quán)重，-1表示不改變?cè)嫉臋?quán)重        debug=False))                         # debug模式test_cfg = dict(    rpn=dict(                                 # 推斷時(shí)的RPN參數(shù)        nms_across_levels=False,              # 在所有的fpn層內(nèi)做nms        nms_pre=2000,                         # 在nms之前保留的的得分最高的proposal數(shù)量        nms_post=2000,                        # 在nms之后保留的的得分最高的proposal數(shù)量        max_num=2000,                         # 在后處理完成之后保留的proposal數(shù)量        nms_thr=0.7,                          # nms閾值        min_bbox_size=0),                     # 最小bbox尺寸    rcnn=dict(        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100)   # max_per_img表示最終輸出的det bbox數(shù)量    # soft-nms is also supported for rcnn testing    # e.g., nms=dict(type='soft_nms', iou_thr=0.5, min_score=0.05)            # soft_nms參數(shù))# dataset settingsdataset_type = 'CocoDataset'                # 數(shù)據(jù)集類(lèi)型data_root = 'data/coco/'                    # 數(shù)據(jù)集根目錄img_norm_cfg = dict(    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)   # 輸入圖像初始化，減去均值mean并處以方差std，to_rgb表示將bgr轉(zhuǎn)為rgbdata = dict(    imgs_per_gpu=2,                # 每個(gè)gpu計(jì)算的圖像數(shù)量    workers_per_gpu=2,             # 每個(gè)gpu分配的線程數(shù)    train=dict(        type=dataset_type,                                                 # 數(shù)據(jù)集類(lèi)型        ann_file=data_root + 'annotations/instances_train2017.json',       # 數(shù)據(jù)集annotation路徑        img_prefix=data_root + 'train2017/',                               # 數(shù)據(jù)集的圖片路徑        img_scale=(1333, 800),                                             # 輸入圖像尺寸，最大邊1333，最小邊800        img_norm_cfg=img_norm_cfg,                                         # 圖像初始化參數(shù)        size_divisor=32,                                                   # 對(duì)圖像進(jìn)行resize時(shí)的最小單位，32表示所有的圖像都會(huì)被resize成32的倍數(shù)        flip_ratio=0.5,                                                    # 圖像的隨機(jī)左右翻轉(zhuǎn)的概率        with_mask=False,                                                   # 訓(xùn)練時(shí)附帶mask        with_crowd=True,                                                   # 訓(xùn)練時(shí)附帶difficult的樣本        with_label=True),                                                  # 訓(xùn)練時(shí)附帶label    val=dict(        type=dataset_type,                                                 # 同上        ann_file=data_root + 'annotations/instances_val2017.json',         # 同上        img_prefix=data_root + 'val2017/',                                 # 同上        img_scale=(1333, 800),                                             # 同上        img_norm_cfg=img_norm_cfg,                                         # 同上        size_divisor=32,                                                   # 同上        flip_ratio=0,                                                      # 同上        with_mask=False,                                                   # 同上        with_crowd=True,                                                   # 同上        with_label=True),                                                  # 同上    test=dict(        type=dataset_type,                                                 # 同上        ann_file=data_root + 'annotations/instances_val2017.json',         # 同上        img_prefix=data_root + 'val2017/',                                 # 同上        img_scale=(1333, 800),                                             # 同上        img_norm_cfg=img_norm_cfg,                                         # 同上        size_divisor=32,                                                   # 同上        flip_ratio=0,                                                      # 同上        with_mask=False,                                                   # 同上        with_label=False,                                                  # 同上        test_mode=True))                                                   # 同上# optimizeroptimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)   # 優(yōu)化參數(shù)，lr為學(xué)習(xí)率，momentum為動(dòng)量因子，weight_decay為權(quán)重衰減因子optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))          # 梯度均衡參數(shù)# learning policylr_config = dict(    policy='step',                        # 優(yōu)化策略    warmup='linear',                      # 初始的學(xué)習(xí)率增加的策略，linear為線性增加    warmup_iters=500,                     # 在初始的500次迭代中學(xué)習(xí)率逐漸增加    warmup_ratio=1.0 / 3,                 # 起始的學(xué)習(xí)率    step=[8, 11])                         # 在第8和11個(gè)epoch時(shí)降低學(xué)習(xí)率checkpoint_config = dict(interval=1)      # 每1個(gè)epoch存儲(chǔ)一次模型# yapf:disablelog_config = dict(    interval=50,                          # 每50個(gè)batch輸出一次信息    hooks=[        dict(type='TextLoggerHook'),      # 控制臺(tái)輸出信息的風(fēng)格        # dict(type='TensorboardLoggerHook')    ])# yapf:enable# runtime settingstotal_epochs = 12                               # 最大epoch數(shù)dist_params = dict(backend='nccl')              # 分布式參數(shù)log_level = 'INFO'                              # 輸出信息的完整度級(jí)別work_dir = './work_dirs/faster_rcnn_r50_fpn_1x' # log文件和模型文件存儲(chǔ)路徑load_from = None                                # 加載模型的路徑，None表示從預(yù)訓(xùn)練模型加載resume_from = None                              # 恢復(fù)訓(xùn)練模型的路徑workflow = [('train', 1)]                       # 當(dāng)前工作區(qū)名稱(chēng)

# model settingsmodel = dict(    type='CascadeRCNN',    num_stages=3,                     # RCNN網(wǎng)絡(luò)的stage數(shù)量，在faster-RCNN中為1    pretrained='modelzoo://resnet50',    backbone=dict(        type='ResNet',        depth=50,        num_stages=4,        out_indices=(0, 1, 2, 3),        frozen_stages=1,        style='pytorch'),    neck=dict(        type='FPN',        in_channels=[256, 512, 1024, 2048],        out_channels=256,        num_outs=5),    rpn_head=dict(        type='RPNHead',        in_channels=256,        feat_channels=256,        anchor_scales=[8],        anchor_ratios=[0.5, 1.0, 2.0],        anchor_strides=[4, 8, 16, 32, 64],        target_means=[.0, .0, .0, .0],        target_stds=[1.0, 1.0, 1.0, 1.0],        use_sigmoid_cls=True),    bbox_roi_extractor=dict(        type='SingleRoIExtractor',        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),        out_channels=256,        featmap_strides=[4, 8, 16, 32]),    bbox_head=[        dict(            type='SharedFCBBoxHead',            num_fcs=2,            in_channels=256,            fc_out_channels=1024,            roi_feat_size=7,            num_classes=81,            target_means=[0., 0., 0., 0.],            target_stds=[0.1, 0.1, 0.2, 0.2],            reg_class_agnostic=True),        dict(            type='SharedFCBBoxHead',            num_fcs=2,            in_channels=256,            fc_out_channels=1024,            roi_feat_size=7,            num_classes=81,            target_means=[0., 0., 0., 0.],            target_stds=[0.05, 0.05, 0.1, 0.1],            reg_class_agnostic=True),        dict(            type='SharedFCBBoxHead',            num_fcs=2,            in_channels=256,            fc_out_channels=1024,            roi_feat_size=7,            num_classes=81,            target_means=[0., 0., 0., 0.],            target_stds=[0.033, 0.033, 0.067, 0.067],            reg_class_agnostic=True)    ])# model training and testing settingstrain_cfg = dict(    rpn=dict(        assigner=dict(            type='MaxIoUAssigner',            pos_iou_thr=0.7,            neg_iou_thr=0.3,            min_pos_iou=0.3,            ignore_iof_thr=-1),        sampler=dict(            type='RandomSampler',            num=256,            pos_fraction=0.5,            neg_pos_ub=-1,            add_gt_as_proposals=False),        allowed_border=0,        pos_weight=-1,        smoothl1_beta=1 / 9.0,        debug=False),    rcnn=[                    # 注意，這里有3個(gè)RCNN的模塊，對(duì)應(yīng)開(kāi)頭的那個(gè)RCNN的stage數(shù)量        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.5,                neg_iou_thr=0.5,                min_pos_iou=0.5,                ignore_iof_thr=-1),            sampler=dict(                type='RandomSampler',                num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False),        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.6,                neg_iou_thr=0.6,                min_pos_iou=0.6,                ignore_iof_thr=-1),            sampler=dict(                type='RandomSampler',                num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False),        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.7,                neg_iou_thr=0.7,                min_pos_iou=0.7,                ignore_iof_thr=-1),            sampler=dict(                type='RandomSampler',                num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False)    ],    stage_loss_weights=[1, 0.5, 0.25])     # 3個(gè)RCNN的stage的loss權(quán)重test_cfg = dict(    rpn=dict(        nms_across_levels=False,        nms_pre=2000,        nms_post=2000,        max_num=2000,        nms_thr=0.7,        min_bbox_size=0),    rcnn=dict(        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=100),    keep_all_stages=False)         # 是否保留所有stage的結(jié)果# dataset settingsdataset_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)data = dict(    imgs_per_gpu=2,    workers_per_gpu=2,    train=dict(        type=dataset_type,        ann_file=data_root + 'annotations/instances_train2017.json',        img_prefix=data_root + 'train2017/',        img_scale=(1333, 800),        img_norm_cfg=img_norm_cfg,        size_divisor=32,        flip_ratio=0.5,        with_mask=False,        with_crowd=True,        with_label=True),    val=dict(        type=dataset_type,        ann_file=data_root + 'annotations/instances_val2017.json',        img_prefix=data_root + 'val2017/',        img_scale=(1333, 800),        img_norm_cfg=img_norm_cfg,        size_divisor=32,        flip_ratio=0,        with_mask=False,        with_crowd=True,        with_label=True),    test=dict(        type=dataset_type,        ann_file=data_root + 'annotations/instances_val2017.json',        img_prefix=data_root + 'val2017/',        img_scale=(1333, 800),        img_norm_cfg=img_norm_cfg,        size_divisor=32,        flip_ratio=0,        with_mask=False,        with_label=False,        test_mode=True))# optimizeroptimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))# learning policylr_config = dict(    policy='step',    warmup='linear',    warmup_iters=500,    warmup_ratio=1.0 / 3,    step=[8, 11])checkpoint_config = dict(interval=1)# yapf:disablelog_config = dict(    interval=50,    hooks=[        dict(type='TextLoggerHook'),        # dict(type='TensorboardLoggerHook')    ])# yapf:enable# runtime settingstotal_epochs = 12dist_params = dict(backend='nccl')log_level = 'INFO'work_dir = './work_dirs/cascade_rcnn_r50_fpn_1x'load_from = Noneresume_from = Noneworkflow = [('train', 1)]

# fp16 settingsfp16 = dict(loss_scale=512.)# model settingsmodel = dict(    type='CascadeRCNN',    num_stages=3,    pretrained='torchvision://resnet50',    backbone=dict(        type='ResNet',        depth=50,        num_stages=4,        out_indices=(0, 1, 2, 3),        frozen_stages=1,        style='pytorch',        #dcn=dict( #在最后三個(gè)block加入可變形卷積          #   modulated=False, deformable_groups=1, fallback_on_stride=False),          #  stage_with_dcn=(False, True, True, True)        ),    neck=dict(        type='FPN',        in_channels=[256, 512, 1024, 2048],        out_channels=256,        num_outs=5),    rpn_head=dict(        type='RPNHead',        in_channels=256,        feat_channels=256,        anchor_scales=[8],        anchor_ratios=[0.2, 0.5, 1.0, 2.0, 5.0], # 添加了0.2，5，過(guò)兩天發(fā)圖        anchor_strides=[4, 8, 16, 32, 64],        target_means=[.0, .0, .0, .0],        target_stds=[1.0, 1.0, 1.0, 1.0],        loss_cls=dict(            type='FocalLoss', use_sigmoid=True, loss_weight=1.0), # 修改了loss，為了調(diào)控難易樣本與正負(fù)樣本比例        loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),    bbox_roi_extractor=dict(        type='SingleRoIExtractor',        roi_layer=dict(type='RoIAlign', out_size=7, sample_num=2),        out_channels=256,        featmap_strides=[4, 8, 16, 32]),    bbox_head=[        dict(            type='SharedFCBBoxHead',            num_fcs=2,            in_channels=256,            fc_out_channels=1024,            roi_feat_size=7,            num_classes=11,            target_means=[0., 0., 0., 0.],            target_stds=[0.1, 0.1, 0.2, 0.2],            reg_class_agnostic=True,            loss_cls=dict(                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),            loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),        dict(            type='SharedFCBBoxHead',            num_fcs=2,            in_channels=256,            fc_out_channels=1024,            roi_feat_size=7,            num_classes=11,            target_means=[0., 0., 0., 0.],            target_stds=[0.05, 0.05, 0.1, 0.1],            reg_class_agnostic=True,            loss_cls=dict(                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),            loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0)),        dict(            type='SharedFCBBoxHead',            num_fcs=2,            in_channels=256,            fc_out_channels=1024,            roi_feat_size=7,            num_classes=11,            target_means=[0., 0., 0., 0.],            target_stds=[0.033, 0.033, 0.067, 0.067],            reg_class_agnostic=True,            loss_cls=dict(                type='CrossEntropyLoss', use_sigmoid=False, loss_weight=1.0),            loss_bbox=dict(type='SmoothL1Loss', beta=1.0, loss_weight=1.0))    ])# model training and testing settingstrain_cfg = dict(    rpn=dict(        assigner=dict(            type='MaxIoUAssigner',            pos_iou_thr=0.7,            neg_iou_thr=0.3,            min_pos_iou=0.3,            ignore_iof_thr=-1),        sampler=dict(            type='RandomSampler',             num=256,            pos_fraction=0.5,            neg_pos_ub=-1,            add_gt_as_proposals=False),        allowed_border=0,        pos_weight=-1,        debug=False),    rpn_proposal=dict(        nms_across_levels=False,        nms_pre=2000,        nms_post=2000,        max_num=2000,        nms_thr=0.7,        min_bbox_size=0),    rcnn=[        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.4, # 更換                neg_iou_thr=0.4,                min_pos_iou=0.4,                ignore_iof_thr=-1),            sampler=dict(                type='OHEMSampler',                num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False),        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.5,                neg_iou_thr=0.5,                min_pos_iou=0.5,                ignore_iof_thr=-1),            sampler=dict(                type='OHEMSampler', # 解決難易樣本，也解決了正負(fù)樣本比例問(wèn)題。num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False),        dict(            assigner=dict(                type='MaxIoUAssigner',                pos_iou_thr=0.6,                neg_iou_thr=0.6,                min_pos_iou=0.6,                ignore_iof_thr=-1),            sampler=dict(                type='OHEMSampler',                num=512,                pos_fraction=0.25,                neg_pos_ub=-1,                add_gt_as_proposals=True),            pos_weight=-1,            debug=False)    ],    stage_loss_weights=[1, 0.5, 0.25])test_cfg = dict(    rpn=dict(        nms_across_levels=False,        nms_pre=1000,        nms_post=1000,        max_num=1000,        nms_thr=0.7,        min_bbox_size=0),    rcnn=dict(        score_thr=0.05, nms=dict(type='nms', iou_thr=0.5), max_per_img=20)) # 這里可以換為sof_tnms# dataset settingsdataset_type = 'CocoDataset'data_root = '../../data/chongqing1_round1_train1_20191223/'img_norm_cfg = dict(    mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)train_pipeline = [    dict(type='LoadImageFromFile'),    dict(type='LoadAnnotations', with_bbox=True),    dict(type='Resize', img_scale=(492,658), keep_ratio=True), #這里可以更換多尺度[(),()]    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=(492,658),        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(    imgs_per_gpu=8, # 有的同學(xué)不知道batchsize在哪修改，其實(shí)就是修改這里，每個(gè)gpu同時(shí)處理的images數(shù)目。workers_per_gpu=2,    train=dict(        type=dataset_type,        ann_file=data_root + 'fixed_annotations.json', # 更換自己的json文件        img_prefix=data_root + 'images/', # images目錄        pipeline=train_pipeline),    val=dict(        type=dataset_type,        ann_file=data_root + 'fixed_annotations.json',        img_prefix=data_root + 'images/',        pipeline=test_pipeline),    test=dict(        type=dataset_type,        ann_file=data_root + 'fixed_annotations.json',        img_prefix=data_root + 'images/',        pipeline=test_pipeline))# optimizeroptimizer = dict(type='SGD', lr=0.001, momentum=0.9, weight_decay=0.0001) # lr = 0.00125*batch_size，不能過(guò)大，否則梯度爆炸。optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2))# learning policylr_config = dict(    policy='step',    warmup='linear',    warmup_iters=500,    warmup_ratio=1.0 / 3,    step=[6, 12, 19])checkpoint_config = dict(interval=1)# yapf:disablelog_config = dict(    interval=64,    hooks=[        dict(type='TextLoggerHook'), # 控制臺(tái)輸出信息的風(fēng)格        # dict(type='TensorboardLoggerHook') # 需要安裝tensorflow and tensorboard才可以使用    ])# yapf:enable# runtime settingstotal_epochs = 20dist_params = dict(backend='nccl')log_level = 'INFO'work_dir = '../work_dirs/cascade_rcnn_r50_fpn_1x' # 日志目錄load_from = '../work_dirs/cascade_rcnn_r50_fpn_1x/latest.pth' # 模型加載目錄文件#load_from = '../work_dirs/cascade_rcnn_r50_fpn_1x/cascade_rcnn_r50_coco_pretrained_weights_classes_11.pth'resume_from = Noneworkflow = [('train', 1)]