Model card for convnextv2-nano-k5-in1k-128

ConvNeXtV2-Nano (k=5) trained from scratch on ImageNet-1K @ 128x128.

Training Recipe:

• ConvNeXt-V2 (Woo et al., 2023) Table 9, IN-1K Nano end-to-end.
• BF16 AMP

Model Details

• Model: convnextv2_nano
• Model Kwargs:
  • kernel_sizes: 5
• Input Size:
  • 3x128x128
• Top-1 Accuracy: 76.71000001876831

Model Usage

import timm
model = timm.create_model('hf-hub:PRadecki/convnextv2-nano-k5-in1k-128', pretrained=True)
model.eval()