Model card for convnextv2-base-k3-silu-gn-in1k-128

ConvNeXtV2-Base (k=3) trained from scratch on ImageNet-1K @128x128.

Training Recipe:

• ConvNeXt supervised recipe -- Table 5.
• BF16 AMP

Model Details

• Model: convnextv2_base
• Model Kwargs:
  • act_layer: silu
  • conv_mlp: True
  • kernel_sizes: 3
  • norm_layer: groupnorm
• Input Size:
  • 3x128x128
• Top-1 Accuracy: 82.12200002899169

Model Usage

import timm
model = timm.create_model('hf-hub:PRadecki/convnextv2-base-k3-silu-gn-128-in1k', pretrained=True)
model.eval()