dior-h测试效果很低

[JiakangSun1]

作者您好，我在训练Dior-H时用的是train和val，训练到后面在trainval上面的准确率挺高的

但是在测试集上的效果却很差

请问是什么原因呢，求作者大人指教
auto_scale_lr = dict(base_batch_size=16, enable=False) backend_args = None crop_size = ( 800, 800, ) data_root = 'F:\\Remote_Sensing\\dior' dataset_type = 'DIORDataset' default_hooks = dict( checkpoint=dict(interval=3, type='CheckpointHook'), logger=dict(interval=50, type='LoggerHook'), param_scheduler=dict(type='ParamSchedulerHook'), sampler_seed=dict(type='DistSamplerSeedHook'), timer=dict(type='IterTimerHook'), visualization=dict(type='DetVisualizationHook')) default_scope = 'mmdet' env_cfg = dict( cudnn_benchmark=False, dist_cfg=dict(backend='nccl'), mp_cfg=dict(mp_start_method='fork', opencv_num_threads=0)) launcher = 'none' load_from = None log_level = 'INFO' log_processor = dict(by_epoch=True, type='LogProcessor', window_size=50) model = dict( backbone=dict( attn_drop_rate=0.0, depth=12, drop_path_rate=0.3, drop_rate=0.0, embed_dim=768, img_size=800, mlp_ratio=4, num_heads=12, out_indices=[ 3, 5, 7, 11, ], patch_size=16, pretrained='F:\\Remote_Sensing\\vit-b-checkpoint-1599.pth', qk_scale=None, qkv_bias=True, type='RVSA_MTP', use_abs_pos_emb=True, use_checkpoint=True), data_preprocessor=dict( bgr_to_rgb=True, mean=[ 123.675, 116.28, 103.53, ], pad_size_divisor=32, std=[ 58.395, 57.12, 57.375, ], type='DetDataPreprocessor'), neck=dict( in_channels=[ 768, 768, 768, 768, ], num_outs=5, out_channels=256, type='FPN'), roi_head=dict( bbox_head=dict( bbox_coder=dict( target_means=[ 0.0, 0.0, 0.0, 0.0, ], target_stds=[ 0.1, 0.1, 0.2, 0.2, ], type='DeltaXYWHBBoxCoder'), fc_out_channels=1024, in_channels=256, loss_bbox=dict(loss_weight=1.0, type='L1Loss'), loss_cls=dict( loss_weight=1.0, type='CrossEntropyLoss', use_sigmoid=False), num_classes=20, reg_class_agnostic=False, roi_feat_size=7, type='Shared2FCBBoxHead'), bbox_roi_extractor=dict( featmap_strides=[ 4, 8, 16, 32, ], out_channels=256, roi_layer=dict(output_size=7, sampling_ratio=0, type='RoIAlign'), type='SingleRoIExtractor'), type='StandardRoIHead'), rpn_head=dict( anchor_generator=dict( ratios=[ 0.5, 1.0, 2.0, ], scales=[ 8, ], strides=[ 4, 8, 16, 32, 64, ], type='AnchorGenerator'), bbox_coder=dict( target_means=[ 0.0, 0.0, 0.0, 0.0, ], target_stds=[ 1.0, 1.0, 1.0, 1.0, ], type='DeltaXYWHBBoxCoder'), feat_channels=256, in_channels=256, loss_bbox=dict(loss_weight=1.0, type='L1Loss'), loss_cls=dict( loss_weight=1.0, type='CrossEntropyLoss', use_sigmoid=True), type='RPNHead'), test_cfg=dict( rcnn=dict( max_per_img=100, nms=dict(iou_threshold=0.5, type='nms'), score_thr=0.05), rpn=dict( max_per_img=1000, min_bbox_size=0, nms=dict(iou_threshold=0.7, type='nms'), nms_pre=1000)), train_cfg=dict( rcnn=dict( assigner=dict( ignore_iof_thr=-1, match_low_quality=False, min_pos_iou=0.5, neg_iou_thr=0.5, pos_iou_thr=0.5, type='MaxIoUAssigner'), debug=False, pos_weight=-1, sampler=dict( add_gt_as_proposals=True, neg_pos_ub=-1, num=512, pos_fraction=0.25, type='RandomSampler')), rpn=dict( allowed_border=-1, assigner=dict( ignore_iof_thr=-1, match_low_quality=True, min_pos_iou=0.3, neg_iou_thr=0.3, pos_iou_thr=0.7, type='MaxIoUAssigner'), debug=False, pos_weight=-1, sampler=dict( add_gt_as_proposals=False, neg_pos_ub=-1, num=256, pos_fraction=0.5, type='RandomSampler')), rpn_proposal=dict( max_per_img=1000, min_bbox_size=0, nms=dict(iou_threshold=0.7, type='nms'), nms_pre=2000)), type='FasterRCNN') model_wrapper = dict( detect_anomalous_params=False, find_unused_parameters=False, type='MMDistributedDataParallel') optim_wrapper = dict( constructor='LayerDecayOptimizerConstructor_ViT', optimizer=dict( betas=( 0.9, 0.999, ), lr=0.0001, type='AdamW', weight_decay=0.05), paramwise_cfg=dict(layer_decay_rate=0.9, num_layers=12)) param_scheduler = [ dict( begin=0, by_epoch=False, end=500, start_factor=1e-06, type='LinearLR'), dict( begin=0, by_epoch=True, end=12, gamma=0.1, milestones=[ 8, 11, ], type='MultiStepLR'), ] resume = False test_cfg = dict(type='TestLoop') test_dataloader = dict( batch_size=1, dataset=dict( ann_file='DIOR_test_coco.json', backend_args=None, data_prefix=dict(img='JPEGImages-test'), data_root='F:\\Remote_Sensing\\dior', pipeline=[ dict(backend_args=None, type='LoadImageFromFile'), dict(keep_ratio=True, scale=( 800, 800, ), type='Resize'), dict(type='LoadAnnotations', with_bbox=True), dict( meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor', ), type='PackDetInputs'), ], test_mode=True, type='DIORDataset'), drop_last=False, num_workers=8, persistent_workers=True, sampler=dict(shuffle=False, type='DefaultSampler')) test_evaluator = dict( ann_file='F:\\Remote_Sensing\\dior/DIOR_test_coco.json', backend_args=None, format_only=False, metric='bbox', type='CocoMetric') test_pipeline = [ dict(backend_args=None, type='LoadImageFromFile'), dict(keep_ratio=True, scale=( 800, 800, ), type='Resize'), dict(type='LoadAnnotations', with_bbox=True), dict( meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor', ), type='PackDetInputs'), ] train_cfg = dict(max_epochs=12, type='EpochBasedTrainLoop', val_interval=3) train_dataloader = dict( batch_sampler=dict(type='AspectRatioBatchSampler'), batch_size=4, dataset=dict( ann_file='DIOR_trainval_coco.json', backend_args=None, data_prefix=dict(img='JPEGImages-trainval'), data_root='F:\\Remote_Sensing\\dior', filter_cfg=dict(filter_empty_gt=True, min_size=32), pipeline=[ dict(backend_args=None, type='LoadImageFromFile'), dict(type='LoadAnnotations', with_bbox=True), dict(keep_ratio=True, scale=( 800, 800, ), type='Resize'), dict(prob=0.5, type='RandomFlip'), dict(type='PackDetInputs'), ], type='DIORDataset'), num_workers=4, persistent_workers=True, sampler=dict(shuffle=True, type='DefaultSampler')) train_pipeline = [ dict(backend_args=None, type='LoadImageFromFile'), dict(type='LoadAnnotations', with_bbox=True), dict(keep_ratio=True, scale=( 800, 800, ), type='Resize'), dict(prob=0.5, type='RandomFlip'), dict(type='PackDetInputs'), ] val_cfg = dict(type='ValLoop') val_dataloader = dict( batch_size=1, dataset=dict( ann_file='DIOR_test_coco.json', backend_args=None, data_prefix=dict(img='JPEGImages-test'), data_root='F:\\Remote_Sensing\\dior', pipeline=[ dict(backend_args=None, type='LoadImageFromFile'), dict(keep_ratio=True, scale=( 800, 800, ), type='Resize'), dict(type='LoadAnnotations', with_bbox=True), dict( meta_keys=( 'img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor', ), type='PackDetInputs'), ], test_mode=True, type='DIORDataset'), drop_last=False, num_workers=8, persistent_workers=True, sampler=dict(shuffle=False, type='DefaultSampler')) val_evaluator = dict( ann_file='F:\\Remote_Sensing\\dior/DIOR_test_coco.json', backend_args=None, format_only=False, metric='bbox', type='CocoMetric') vis_backends = [ dict(type='LocalVisBackend'), ] visualizer = dict( name='visualizer', type='DetLocalVisualizer', vis_backends=[ dict(type='LocalVisBackend'), ]) work_dir = './work_dirs\\dior'

[DotWang]

@JiakangSun1 你这个图我看不出来准确率高，你看看我那些log里边是啥样的

[JiakangSun1]

@DotWang 感谢作者的回复，之前配置文件里model写成了RVSA_MTP，现在改成RVSA_MTP_branches已经能够正常训练了