Model Card for nextvit_s_eu-common

A Next-ViT small image classification model. This model was trained on the eu-common dataset containing common European bird species.

The species list is derived from the Collins bird guide [^1].

[^1]: Svensson, L., Mullarney, K., & Zetterström, D. (2022). Collins bird guide (3rd ed.). London, England: William Collins.

Note: A 256 x 256 variant of this model is available as nextvit_s_eu-common256px.

Model Details

• Model Type: Image classification and detection backbone

• Model Stats:

  • Params (M): 31.5
  • Input image size: 384 x 384

• Dataset: eu-common (707 classes)

• Papers:

  • Next-ViT: Next Generation Vision Transformer for Efficient Deployment in Realistic Industrial Scenarios: https://arxiv.org/abs/2207.05501

Model Usage

Image Classification

import birder
from birder.inference.classification import infer_image

(net, model_info) = birder.load_pretrained_model("nextvit_s_eu-common", inference=True)
# Note: A 256x256 variant is available as "nextvit_s_eu-common256px"

# Get the image size the model was trained on
size = birder.get_size_from_signature(model_info.signature)

# Create an inference transform
transform = birder.classification_transform(size, model_info.rgb_stats)

image = "path/to/image.jpeg"  # or a PIL image, must be loaded in RGB format
(out, _) = infer_image(net, image, transform)
# out is a NumPy array with shape of (1, 707), representing class probabilities.

Image Embeddings

import birder
from birder.inference.classification import infer_image

(net, model_info) = birder.load_pretrained_model("nextvit_s_eu-common", inference=True)

# Get the image size the model was trained on
size = birder.get_size_from_signature(model_info.signature)

# Create an inference transform
transform = birder.classification_transform(size, model_info.rgb_stats)

image = "path/to/image.jpeg"  # or a PIL image
(out, embedding) = infer_image(net, image, transform, return_embedding=True)
# embedding is a NumPy array with shape of (1, 1024)

Detection Feature Map

from PIL import Image
import birder

(net, model_info) = birder.load_pretrained_model("nextvit_s_eu-common", inference=True)

# Get the image size the model was trained on
size = birder.get_size_from_signature(model_info.signature)

# Create an inference transform
transform = birder.classification_transform(size, model_info.rgb_stats)

image = Image.open("path/to/image.jpeg")
features = net.detection_features(transform(image).unsqueeze(0))
# features is a dict (stage name -> torch.Tensor)
print([(k, v.size()) for k, v in features.items()])
# Output example:
# [('stage1', torch.Size([1, 96, 96, 96])),
#  ('stage2', torch.Size([1, 256, 48, 48])),
#  ('stage3', torch.Size([1, 512, 24, 24])),
#  ('stage4', torch.Size([1, 1024, 12, 12]))]

Citation

@misc{li2022nextvitgenerationvisiontransformer,
      title={Next-ViT: Next Generation Vision Transformer for Efficient Deployment in Realistic Industrial Scenarios},
      author={Jiashi Li and Xin Xia and Wei Li and Huixia Li and Xing Wang and Xuefeng Xiao and Rui Wang and Min Zheng and Xin Pan},
      year={2022},
      eprint={2207.05501},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2207.05501},
}