How to decrease the input image size??

[leesangjoon1]

In detectron2, the config file has the input_size control code before resize the input_size

but how to do that in mmdet??

Due to the memory cost, now I am having trouble in training the deep learing.

Please help me :)

[jbwang1997]

You can modify the img_scale in Resize pipeline to change the input_size.

[leesangjoon1]

You can modify the img_scale in Resize pipeline to change the input_size.

yeah I resize the input image size but I can run the training that size in detectron2 without out of memory but in mmdet it become out of memory so I don't what is the problem.

[jbwang1997]

Could you provide pipeline you used.

[leesangjoon1]

Could you provide pipeline you used.

This is coco_detection.py file In that file, the input_size is 1333, 800 but in detectron2 I train input_size 2500, 2500 So I don't know what do I change And I heard that the samples_per_gpu mean Batch_size. But I tried the batch size 32 in detectron. but it alos occur out of memory. Please help me

dataset_type = 'CocoDataset' data_root = '/home/sangjoon/Yet-Another-EfficientDet-Pytorch/datasets/coco_white/' 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=(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']), ]) ] data = dict( samples_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/', pipeline=train_pipeline), val=dict( type=dataset_type, ann_file=data_root + 'annotations/instances_val2017.json', img_prefix=data_root + 'val2017/', pipeline=test_pipeline), test=dict( type=dataset_type, ann_file=data_root + 'annotations/instances_val2017.json', img_prefix=data_root + 'val2017/', pipeline=test_pipeline)) evaluation = dict(interval=1, metric='bbox')

[jbwang1997]

According to the previous comparison, the training memory of mmdetection and detectron2 are basically identical. Please check if the running conditions are the same.

[jbwang1997]

For out of memory. You can reduce the sample_per_gpu and further reduce the input image size.

[jbwang1997]

This issue will be closed due to inactivity. Feel free to reopen it if you have any questions.