efficientformerv2_l.snap_dist_in1k

Clasificación de Imágenes

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

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

model = timm.create_model('efficientformerv2_l.snap_dist_in1k', pretrained=True)
model = model.eval()

# obtener transformaciones específicas del modelo (normalización, cambio de tamaño)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0)) # añadir imagen única en lote de 1

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

Embeddings de Imágenes

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

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

model = timm.create_model(
'efficientformerv2_l.snap_dist_in1k',
pretrained=True,
num_classes=0, # eliminar clasificador nn.Linear
)
model = model.eval()

# obtener transformaciones específicas del modelo (normalización, cambio de tamaño)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0)) # el output es un tensor con forma (batch_size, num_features)

# o de manera equivalente (sin necesidad de establecer num_classes=0)

output = model.forward_features(transforms(img).unsqueeze(0))
# el output no está agrupado (es decir, un tensor con forma (batch_size, num_features, H, W))

output = model.forward_head(output, pre_logits=True)
# el output es un tensor con forma (batch_size, num_features)

Extracción de Mapas de Características

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

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

model = timm.create_model(
'efficientformerv2_l.snap_dist_in1k',
pretrained=True,
features_only=True,
)
model = model.eval()

# obtener transformaciones específicas del modelo (normalización, cambio de tamaño)
data_config = timm.data.resolve_model_data_config(model)
transforms = timm.data.create_transform(**data_config, is_training=False)

output = model(transforms(img).unsqueeze(0)) # añadir imagen única en lote de 1

for o in output:
# imprimir forma de cada mapa de características en el output
# e.g. para efficientformerv2_l:
# torch.Size([2, 40, 56, 56])
# torch.Size([2, 80, 28, 28])
# torch.Size([2, 192, 14, 14])
# torch.Size([2, 384, 7, 7])
print(o.shape)

Comparación de Modelos

efficientformerv2_l.snap_dist_in1k

• top1: 83.628
• top5: 96.54
• param_count: 26.32
• img_size: 224

efficientformer_l7.snap_dist_in1k

• top1: 83.368
• top5: 96.534
• param_count: 82.23
• img_size: 224

efficientformer_l3.snap_dist_in1k

• top1: 82.572
• top5: 96.24
• param_count: 31.41
• img_size: 224

efficientformerv2_s2.snap_dist_in1k

• top1: 82.128
• top5: 95.902
• param_count: 12.71
• img_size: 224

efficientformer_l1.snap_dist_in1k

• top1: 80.496
• top5: 94.984
• param_count: 12.29
• img_size: 224

efficientformerv2_s1.snap_dist_in1k

• top1: 79.698
• top5: 94.698
• param_count: 6.19
• img_size: 224

efficientformerv2_s0.snap_dist_in1k

• top1: 76.026
• top5: 92.77
• param_count: 3.6
• img_size: 224