model size was not changed by reduce the crop size and resize

[huang45]

changes in the config file:

crop_size = (512, 512)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations'),
dict(type='Resize', img_scale=(860, 540), ratio_range=(0.5, 2.0)),
dict(type='RandomCrop', crop_size=crop_size, cat_max_ratio=0.75),
dict(type='RandomFlip', flip_ratio=0.5),
dict(type='PhotoMetricDistortion'),
dict(type='Normalize', **img_norm_cfg),
dict(type='Pad', size=crop_size, pad_val=0, seg_pad_val=255),
dict(type='DefaultFormatBundle'),
dict(type='Collect', keys=['img', 'gt_semantic_seg']),
]
test_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(860, 540),
# img_ratios=[0.5, 0.75, 1.0, 1.25, 1.5, 1.75],
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='RandomFlip'),
dict(type='Normalize', **img_norm_cfg),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img']),
])
]

but the model size was not changed.
The input shape was not changed:

Input shape: (3, 2048, 1024)
Flops: 7.53 GFLOPs
Params: 1.21 M

[xvjiarui]

Hi @huang45
Are you using tools/get_flops.py? If so, you may try passing --shape 860 540

[huang45]

@xvjiarui Yes. By adding --shape 860 540, the Flops is reduced to 1.69 GFLOPs, but the Params is same. The pth file size is also not changed. Is it normal? thanks.

Input shape: (3, 860, 540)
Flops: 1.69 GFLOPs
Params: 1.21 M

[xvjiarui]

Hi @huang45
The model size or the number of params is not related to input size.