RuntimeError: mat1 and mat2 shapes cannot be multiplied (458752x7 and 100352x3)

[YusukeImadegawa]

I changed backborn from VGG18 to ResNet101 cuz VGG is not good perfomance. I got the Error RuntimeError: mat1 and mat2 shapes cannot be multiplied (458752x7 and 100352x3) when I fine-tune with setting the below config.
Input : 224*224 RGB image
Output : 3 chanel
I think # is the cause of error, but i can't set proper one.

Config

_base_ = [
    '../_base_/models/resnet101.py', 
    '../_base_/datasets/imagenet_bs256_davit_224.py', 
    '../_base_/schedules/imagenet_bs1024_adamw_conformer.py', 
    '../_base_/default_runtime.py'
]

model = dict(
    type='ImageClassifier',
    backbone=dict(
        type='ResNet',
        depth=101,
        num_stages=4, #
        in_channels=3, #
        out_indices=(3, ), #
        init_cfg=dict( 
            type='Pretrained',
            checkpoint='https://download.openmmlab.com/mmclassification/v0/resnet/resnet101_8xb32_in1k_20210831-539c63f8.pth',
            prefix='backbone',
        ),
    ),
    neck=None, #
    head=dict(
        type='LinearClsHead',
        num_classes=3,
        in_channels=7*7*2048, #
        loss=dict(type='CrossEntropyLoss', loss_weight=1.0),
        topk=(1, ),
        cal_acc=True, 
    ))

Structure

+--------------------------+----------------------+-----------+--------------+
| module | #parameters or shape | #flops | #activations |
+--------------------------+----------------------+-----------+--------------+
| model | 42.801M | 7.832G | 16.232M |
| backbone | 42.5M | 7.832G | 16.232M |
| backbone.conv1 | 9.408K | 0.118G | 0.803M |
| backbone.conv1.weight | (64, 3, 7, 7) | | |
| backbone.bn1 | 0.128K | 1.606M | 0 |
| backbone.bn1.weight | (64,) | | |
| backbone.bn1.bias | (64,) | | |
| backbone.layer1 | 0.216M | 0.677G | 4.415M |
| backbone.layer1.0 | 75.008K | 0.235G | 2.007M |
| backbone.layer1.1 | 70.4K | 0.221G | 1.204M |
| backbone.layer1.2 | 70.4K | 0.221G | 1.204M |
| backbone.layer2 | 1.22M | 1.034G | 3.111M |
| backbone.layer2.0 | 0.379M | 0.375G | 1.305M |
| backbone.layer2.1 | 0.28M | 0.22G | 0.602M |
| backbone.layer2.2 | 0.28M | 0.22G | 0.602M |
| backbone.layer2.3 | 0.28M | 0.22G | 0.602M |
| backbone.layer3 | 26.09M | 5.191G | 7.276M |
| backbone.layer3.0 | 1.512M | 0.374G | 0.652M |
| backbone.layer3.1 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.2 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.3 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.4 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.5 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.6 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.7 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.8 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.9 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.10 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.11 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.12 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.13 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.14 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.15 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.16 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.17 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.18 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.19 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.20 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.21 | 1.117M | 0.219G | 0.301M |
| backbone.layer3.22 | 1.117M | 0.219G | 0.301M |
| backbone.layer4 | 14.965M | 0.81G | 0.627M |
| backbone.layer4.0 | 6.04M | 0.373G | 0.326M |
| backbone.layer4.1 | 4.463M | 0.219G | 0.151M |
| backbone.layer4.2 | 4.463M | 0.219G | 0.151M |
| head.fc | 0.301M | | |
| head.fc.weight | (3, 100352) | | |
| head.fc.bias | (3,) | | |
+--------------------------+----------------------+-----------+--------------+

==============================
Input shape: (3, 224, 224)
Flops: 7.832G
Params: 42.801M
Activation: 16.232M

[Ezra-Yu]

The following is enough fro model setting:

_base_ = [
    '../_base_/models/resnet101.py',           # model settings
     ...............
]

# >>>>>>>>>>>>>>> Override model settings here >>>>>>>>>>>>>>>>>>>
model = dict(
    backbone=dict(
        init_cfg=dict( 
            type='Pretrained',
            checkpoint='https://download.openmmlab.com/mmclassification/v0/resnet/resnet101_8xb32_in1k_20210831-539c63f8.pth',
            prefix='backbone',
        ),
    ),
    head=dict(num_classes=3),
)