ImportError: cannot import name 'convert_splitbn_model' from 'timm.models'

[watertianyi]

Package Version

---

certifi 2024.7.4
charset-normalizer 3.3.2
filelock 3.15.4
fsspec 2024.6.1
fvcore 0.1.5.post20221221
huggingface-hub 0.24.6
idna 3.7
iopath 0.1.10
Jinja2 3.1.4
MarkupSafe 2.1.5
mpmath 1.3.0
networkx 3.1
numpy 1.24.4
nvidia-cublas-cu12 12.1.3.1
nvidia-cuda-cupti-cu12 12.1.105
nvidia-cuda-nvrtc-cu12 12.1.105
nvidia-cuda-runtime-cu12 12.1.105
nvidia-cudnn-cu12 9.1.0.70
nvidia-cufft-cu12 11.0.2.54
nvidia-curand-cu12 10.3.2.106
nvidia-cusolver-cu12 11.4.5.107
nvidia-cusparse-cu12 12.1.0.106
nvidia-nccl-cu12 2.20.5
nvidia-nvjitlink-cu12 12.6.20
nvidia-nvtx-cu12 12.1.105
packaging 24.1
pillow 10.4.0
pip 24.2
portalocker 2.10.1
PyYAML 6.0.2
requests 2.32.3
safetensors 0.4.4
setuptools 72.1.0
sympy 1.13.2
tabulate 0.9.0
termcolor 2.4.0
timm 1.0.8
torch 2.4.0
torchvision 0.19.0
tqdm 4.66.5
triton 3.0.0
typing_extensions 4.12.2
urllib3 2.2.2
wheel 0.43.0
yacs 0.1.8

[Lupin1998]

Hi, @goldwater668. Thanks for using MogaNet. Unfortunately, this error might be caused by timm itself in its new version. I suggest you search (or raise) this issue under the repo of timm, or you could try an early version of timm (e.g., 0.6.11).

[watertianyi]

CUDA_VISIBLE_DEVICES=0 python validate.py
--model moganet_xtiny
--img_size 224
--amp
--batch_size 4
--crop_pct 0.9
--data_dir /val_NG_single_result
--dataset val_NG_Qiepian_all
--checkpoint MogaNet/save_results/model_best.pth.tar
--num_classes 2
--results_file MogaNet/0823_val.csv

Test: [ 0/3761] Time: 15.820s (15.820s, 0.25/s) Loss: 0.1131 (0.1131) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 10/3761] Time: 0.036s (1.471s, 2.72/s) Loss: 0.1017 (0.1100) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 20/3761] Time: 0.037s (0.787s, 5.08/s) Loss: 0.1102 (0.1100) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 30/3761] Time: 0.036s (0.545s, 7.34/s) Loss: 0.1129 (0.1105) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 40/3761] Time: 0.037s (0.421s, 9.50/s) Loss: 0.1063 (0.1100) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 50/3761] Time: 0.035s (0.346s, 11.57/s) Loss: 0.1151 (0.1103) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 60/3761] Time: 0.036s (0.295s, 13.56/s) Loss: 0.1061 (0.1098) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 70/3761] Time: 0.036s (0.259s, 15.47/s) Loss: 0.1130 (0.1099) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 80/3761] Time: 0.036s (0.231s, 17.31/s) Loss: 0.1202 (0.1104) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 90/3761] Time: 0.037s (0.210s, 19.09/s) Loss: 0.1126 (0.1107) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 100/3761] Time: 0.036s (0.192s, 20.80/s) Loss: 0.1031 (0.1106) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 110/3761] Time: 0.038s (0.178s, 22.45/s) Loss: 0.1022 (0.1105) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 120/3761] Time: 0.035s (0.167s, 24.02/s) Loss: 0.1063 (0.1103) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 130/3761] Time: 0.035s (0.157s, 25.54/s) Loss: 0.1071 (0.1102) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 140/3761] Time: 0.037s (0.148s, 27.02/s) Loss: 0.1011 (0.1101) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 150/3761] Time: 0.036s (0.141s, 28.45/s) Loss: 0.1165 (0.1105) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 160/3761] Time: 0.036s (0.134s, 29.83/s) Loss: 0.1087 (0.1106) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 170/3761] Time: 0.034s (0.128s, 31.17/s) Loss: 0.1178 (0.1111) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 180/3761] Time: 0.036s (0.123s, 32.48/s) Loss: 0.1040 (0.1109) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 190/3761] Time: 0.035s (0.119s, 33.73/s) Loss: 0.1123 (0.1105) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 200/3761] Time: 0.036s (0.114s, 34.94/s) Loss: 0.1131 (0.1103) Acc@1: 100.000 (100.000) Acc@5: 100.000 (100.000)
Test: [ 210/3761] Time: 0.036s (0.111s, 36.10/s) Loss: 2.2324 (0.1930) Acc@1: 0.000 ( 96.090) Acc@5: 100.000 (100.000)
Test: [ 220/3761] Time: 0.034s (0.107s, 37.26/s) Loss: 2.1895 (0.2869) Acc@1: 0.000 ( 91.742) Acc@5: 100.000 (100.000)
Test: [ 230/3761] Time: 0.036s (0.104s, 38.37/s) Loss: 2.2363 (0.3717) Acc@1: 0.000 ( 87.771) Acc@5: 100.000 (100.000)
Test: [ 240/3761] Time: 0.035s (0.101s, 39.45/s) Loss: 2.2461 (0.4487) Acc@1: 0.000 ( 84.129) Acc@5: 100.000 (100.000)
Test: [ 250/3761] Time: 0.035s (0.099s, 40.50/s) Loss: 2.4180 (0.5219) Acc@1: 0.000 ( 80.777) Acc@5: 100.000 (100.000)
Test: [ 260/3761] Time: 0.035s (0.096s, 41.53/s) Loss: 2.2832 (0.5877) Acc@1: 0.000 ( 77.682) Acc@5: 100.000 (100.000)
Test: [ 270/3761] Time: 0.034s (0.094s, 42.53/s) Loss: 2.2188 (0.6490) Acc@1: 0.000 ( 74.815) Acc@5: 100.000 (100.000)
Test: [ 280/3761] Time: 0.036s (0.092s, 43.49/s) Loss: 2.2812 (0.7058) Acc@1: 0.000 ( 72.153) Acc@5: 100.000 (100.000)
Test: [ 290/3761] Time: 0.034s (0.090s, 44.44/s) Loss: 2.1133 (0.7576) Acc@1: 0.000 ( 69.674) Acc@5: 100.000 (100.000)
Test: [ 300/3761] Time: 0.037s (0.088s, 45.35/s) Loss: 2.2461 (0.8075) Acc@1: 0.000 ( 67.359) Acc@5: 100.000 (100.000)
../aten/src/ATen/native/cuda/Loss.cu:250: nll_loss_forward_reduce_cuda_kernel_2d: block: [0,0,0], thread: [0,0,0] Assertion t >= 0 && t < n_classes failed.
../aten/src/ATen/native/cuda/Loss.cu:250: nll_loss_forward_reduce_cuda_kernel_2d: block: [0,0,0], thread: [1,0,0] Assertion t >= 0 && t < n_classes failed.
../aten/src/ATen/native/cuda/Loss.cu:250: nll_loss_forward_reduce_cuda_kernel_2d: block: [0,0,0], thread: [2,0,0] Assertion t >= 0 && t < n_classes failed.
../aten/src/ATen/native/cuda/Loss.cu:250: nll_loss_forward_reduce_cuda_kernel_2d: block: [0,0,0], thread: [3,0,0] Assertion t >= 0 && t < n_classes failed.
return torch._C._nn.cross_entropy_loss(input, target, weight, _Reduction.get_enum(reduction), ignore_index, label_smoothing)
RuntimeError: CUDA error: device-side assert triggered
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

[Lupin1998]

Please check your dataset. The provided class number doesn't match the provided label.

[watertianyi]

可以说的具体点吗，数据集目录与命名需要另外标注吗

[Lupin1998]

If you are using a custom dataset, please check these issues by yourself. If you are using ImageNet, please set --num_classes 1000.

[watertianyi]

@Lupin1998 如果已经训练了2个类别，现在新增第三个类别，是否可以增加第三个类别数据进行微调呢

[watertianyi]

@Lupin1998
Can this network be added to a feature pyramid network to be compatible with images of different scales?