fcjian
Please - why can't I reproduce the results
I have downloaded your public datasets and annotations, but I cannot reproduce your experimental results. Could you please advise me? I am looking forward to your reply!
base = [ '../base/models/faster-rcnn_r50_fpn.py', '../base/datasets/coco_detection.py', '../base/schedules/schedule_1x.py', '../base/default_runtime.py' ]
base_data_root = '/gpfsdata/home/yangshuai/open_vocabulary/InstaGen/outputs/coco_ovd_images/coco_base_ft6_1250/' base_train_ann = 'anns_thr0.8_pos_images.json' novel_data_root = '/gpfsdata/home/yangshuai/open_vocabulary/InstaGen/outputs/coco_ovd_images/coco_novel_ft6_3000/' novel_train_ann = 'anns_base_and_novel_thr0.4.json'
coco_data_root='/gpfsdata/home/yangshuai/data/coco/' coco_train_ann='instagen_annotations/instances_train2017_seen.json' test_data_root = '/gpfsdata/home/yangshuai/data/coco/' test_ann = 'annotations/instances_val2017.json'
base_ind_file = '/gpfsdata/home/yangshuai/open_vocabulary/InstaGen/mmdetection/instagen_resources/coco_base_inds.txt' novel_ind_file = '/gpfsdata/home/yangshuai/open_vocabulary/InstaGen/mmdetection/instagen_resources/coco_novel_inds.txt'
model = dict( data_preprocessor=dict( type='DetDataPreprocessor', mean=[103.530, 116.280, 123.675], std=[1.0, 1.0, 1.0], bgr_to_rgb=False, pad_size_divisor=32), backbone=dict( norm_cfg=dict(requires_grad=False), norm_eval=True, style='caffe', init_cfg=dict( type='Pretrained', checkpoint='open-mmlab://detectron2/resnet50_caffe' ) ), train_cfg = dict( rpn_proposal = dict( ignore_loss=True, ignored_data='/gpfsdata/home/yangshuai/data/coco/' ), rcnn = dict( ignore_novel=True, ignored_data='/gpfsdata/home/yangshuai/data/coco/', ) ), roi_head=dict( bbox_head=dict( reg_class_agnostic=False, loss_cls=dict( type='MaskedCrossEntropyLoss', use_sigmoid=True, loss_weight=1.0, num_classes=80, base_ind_file=base_ind_file ), ) ) )
base_dataset=dict( type='CocoDataset', data_root=base_data_root, ann_file=base_train_ann, data_prefix=dict(img='images/'), filter_cfg=dict(filter_empty_gt=True, min_size=32), pipeline=[ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True), dict( type='RandomChoiceResize', scales=[(1333, 224), (1333, 256), (1333, 288), (1333, 320), (1333, 352), (1333, 384), (1333, 416), (1333, 448), (1333, 480), (1333, 512), (1333, 544), (1333, 576), (1333, 608), (1333, 640), (1333, 672), (1333, 704), (1333, 736), (1333, 768), (1333, 800)], keep_ratio=True), dict(type='RandomFlip', prob=0.5), dict(type='PackDetInputs') ], backend_args=None )
novel_dataset=dict( type='CocoDataset', data_root=novel_data_root, ann_file=novel_train_ann, data_prefix=dict(img='images/'), filter_cfg=dict(filter_empty_gt=True, min_size=32), pipeline=[ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True), dict( type='RandomChoiceResize', scales=[(1333, 224), (1333, 256), (1333, 288), (1333, 320), (1333, 352), (1333, 384), (1333, 416), (1333, 448), (1333, 480), (1333, 512), (1333, 544), (1333, 576), (1333, 608), (1333, 640), (1333, 672), (1333, 704), (1333, 736), (1333, 768), (1333, 800)], keep_ratio=True), dict(type='RandomFlip', prob=0.5), dict(type='PackDetInputs') ], backend_args=None )
base_train_dataloader = dict( batch_size=4, num_workers=2, persistent_workers=True, sampler=dict(type='DefaultSampler', shuffle=True), batch_sampler=dict(type='AspectRatioBatchSampler'), dataset=base_dataset, )
novel_train_dataloader = dict( batch_size=4, num_workers=2, persistent_workers=True, sampler=dict(type='DefaultSampler', shuffle=True), batch_sampler=dict(type='AspectRatioBatchSampler'), dataset=novel_dataset, )
coco_train_dataloader = dict( batch_size=4, num_workers=2, persistent_workers=True, sampler=dict(type='DefaultSampler', shuffle=True), batch_sampler=dict(type='AspectRatioBatchSampler'), dataset=dict( type='CocoDataset', data_root=coco_data_root, ann_file=coco_train_ann, data_prefix=dict(img='train2017/'), filter_cfg=dict(filter_empty_gt=True, min_size=32), pipeline=[ dict(type='LoadImageFromFile', backend_args=None), dict(type='LoadAnnotations', with_bbox=True), dict(type='Resize', scale=(1333, 800), keep_ratio=True), dict(type='RandomFlip', prob=0.5), dict(type='PackDetInputs') ], backend_args=None))
test_pipeline=[ dict(type='LoadImageFromFile', backend_args=None), dict(type='Resize', scale=(1333, 800), keep_ratio=True), dict(type='LoadAnnotations', with_bbox=True), dict( type='PackDetInputs', meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'scale_factor')) ]
val_dataloader = dict( dataset=dict( type='CocoDataset', data_root=test_data_root, ann_file=test_ann, data_prefix=dict(img='val2017/'), pipeline=test_pipeline ) )
test_dataloader = dict( dataset=dict( type='CocoDataset', data_root=test_data_root, ann_file=test_ann, data_prefix=dict(img='val2017/'), pipeline=test_pipeline ) )
val_evaluator = dict( type='CocoMetric', ann_file=test_data_root + test_ann, metric='bbox', format_only=False, backend_args=None, classwise=True, base_ind_file=base_ind_file, novel_ind_file=novel_ind_file ) test_evaluator = dict( type='CocoMetric', ann_file=test_data_root + test_ann, metric='bbox', format_only=False, backend_args=None, classwise=True, base_ind_file=base_ind_file, novel_ind_file=novel_ind_file )
train_cfg = dict(type='EpochDatasetTrainLoop', max_epochs=12, val_interval=1,
train_dataloaders=[base_train_dataloader, novel_train_dataloader, coco_train_dataloader],
self_train=False, score_thresh=0., start_epoch=-1)
auto_scale_lr = dict(base_batch_size=16)
work_dir = './work_dirs/faster-rcnn_r50-caffe_c4-1x_coco-ovd_instagen-dataset'
find_unused_parameters = True
@fcjian This is my configuration file, please refer to it, thank you very much!