JaxNN/resnet34.a1_in1k

-- license: apache-2.0 library_name: jaxnn tags: - image-classification

NOTE

jaxnn is still developing (pip installation is not available), it will be available soon when enough models are ported into FLAX/JAX

Model card for resnet34.a1_in1k

A ResNet-B image classification model.

This model features:

• ReLU activations
• single layer 7x7 convolution with pooling
• 1x1 convolution shortcut downsample

Trained on ImageNet-1k in jaxnn using recipe template described below.

Recipe details:

• ResNet Strikes Back A1 recipe
• LAMB optimizer with BCE loss
• Cosine LR schedule with warmup

Model Details

• Model Type: Image classification / feature backbone
• Model Stats:
  • Params (M): 21.8
  • GMACs: 3.7
  • Activations (M): 3.7
  • Image size: train = 224 x 224, test = 288 x 288
• Papers:
  • ResNet strikes back: An improved training procedure in timm: https://arxiv.org/abs/2110.00476
  • Deep Residual Learning for Image Recognition: https://arxiv.org/abs/1512.03385
• Original: https://github.com/huggingface/pytorch-image-models

Model Usage

Image Classification

from urllib.request import urlopen
from PIL import Image
import jaxnn

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = jaxnn.create_model('resnet34.a1_in1k', pretrained=True)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = jaxnn.data.resolve_model_data_config(model)
transforms = jaxnn.data.create_transform(**data_config, is_training=False)

output = model(jax.expand_dims(transforms(img), 0))  # unsqueeze single image into batch of 1

top5_probabilities, top5_class_indices = torch.topk(output.softmax(dim=1) * 100, k=5)

Feature Map Extraction

from urllib.request import urlopen
from PIL import Image
import jaxnn
import jax

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = jaxnn.create_model(
    'resnet34.a1_in1k',
    pretrained=True,
    features_only=True,
)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = jaxnn.data.resolve_model_data_config(model)
transforms = jaxnn.data.create_transform(**data_config, is_training=False)

output = model(jax.expand_dims(transforms(img), 0))  # jax.expand_dims single image into batch of 1

for o in output:
    # print shape of each feature map in output in format [Batch, Height, Width, Channels]
    # e.g.:
    #  (1, 112, 112, 64)
    #  (1, 56, 56, 64)
    #  (1, 28, 28, 128)
    #  (1, 14, 14, 256)
    #  (1, 7, 7, 512)

    print(o.shape)

Image Embeddings

from urllib.request import urlopen
from PIL import Image
import jaxnn

img = Image.open(urlopen(
    'https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/beignets-task-guide.png'
))

model = jaxnn.create_model(
    'resnet34.a1_in1k',
    pretrained=True,
    num_classes=0,  # remove classifier nn.Linear
)
model = model.eval()

# get model specific transforms (normalization, resize)
data_config = jaxnn.data.resolve_model_data_config(model)
transforms = jaxnn.data.create_transform(**data_config, is_training=False)

output = model(jax.expand_dims(transforms(img), 0))   # output is (batch_size, num_features) shaped Array

# or equivalently (without needing to set num_classes=0)

output = model.forward_features(jax.expand_dims(transforms(img), 0))
# output is unpooled, a (1, 7, 7, 512) shaped tensor

output = model.forward_head(output, pre_logits=True)
# output is a (1, num_features) shaped Array