STD with Oriented RCNN and HiViT-B的训练测试与验证集选取问题

[liutianyi00]

您好，我训练了STD with Oriented RCNN and HiViT-B，训练集使用ms_trainval（多尺度），验证集使用ss_trainval（单尺度），测试集使用ms_test，训练18epoch，samples_per_gpu=2，workers_per_gpu=4，以下为配置文件，问什么我最终的结果 COCO style result: AP50: 0.819771851038522 AP75: 0.6116674730023078 mAP: 0.5461242820722623
请问为什么没有达到82.24呢 base = './rotated_imted_hb1_oriented_rcnn_hivitdet_base_1x_dota_le90_16h.py'

model = dict( backbone=dict( use_checkpoint=True, # True, # False for A100 ), roi_head=dict( bbox_head=dict( type='RotatedMAEBBoxHeadSTDC', dc_mode_str_list = ['', '', '', 'XY', '', 'A', '', 'WH'], num_convs_list = [0, 0, 3, 3, 2, 2, 1, 1], am_mode_str_list = ['', '', 'V', 'V', 'V', 'V', 'V', 'V'], rois_mode = 'rbbox', use_checkpoint=True, # True, # False for A100 ), ), )

dota_ms_rr

data_root_ms = 'data/split_ms_dota/' data_root_ss = 'data/split_ss_dota/' angle_version = 'le90' 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='RResize', img_scale=(1024, 1024)), dict( type='RRandomFlip', flip_ratio=[0.25, 0.25, 0.25], direction=['horizontal', 'vertical', 'diagonal'], version=angle_version), dict( type='PolyRandomRotate', rotate_ratio=0.5, angles_range=180, auto_bound=False, rect_classes=[9, 11], version=angle_version), dict(type='Normalize', **img_norm_cfg), dict(type='Pad', size_divisor=32), dict(type='DefaultFormatBundle'), dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels']) ] data = dict( samples_per_gpu=2, workers_per_gpu=4, # 4, # 8 for A100 train=dict( ann_file=data_root_ms + 'trainval/annfiles/', img_prefix=data_root_ms + 'trainval/images/', pipeline=train_pipeline, version=angle_version), val=dict( ann_file=data_root_ss + 'trainval/annfiles/', img_prefix=data_root_ss + 'trainval/images/', version=angle_version), test=dict( ann_file=data_root_ms + 'test/images/', img_prefix=data_root_ms + 'test/images/', version=angle_version))

optimizer

optimizer = dict(lr=5e-5) # 4 GPUs for A100

[yuhongtian17]

AP50=81.98属于正常浮动范围，没有问题。