Image Embedding with Timm

author: Jael Gu, Filip

Description

An image embedding operator generates a vector given an image. This operator extracts features for image with pre-trained models provided by Timm. Timm is a deep-learning library developed by Ross Wightman, who maintains SOTA deep-learning models and tools in computer vision.

Code Example

Load an image from path './towhee.jpeg' and use the pre-trained ResNet50 model ('resnet50') to generate an image embedding.

Write a pipeline with explicit inputs/outputs name specifications:

from towhee import pipe, ops, DataCollection

p = (
    pipe.input('path')
        .map('path', 'img', ops.image_decode())
        .map('img', 'vec', ops.image_embedding.timm(model_name='resnet50'))
        .output('img', 'vec')
)

DataCollection(p('towhee.jpeg')).show()

Factory Constructor

Create the operator via the following factory method:

image_embedding.timm(model_name='resnet34', num_classes=1000, skip_preprocess=False)

Parameters:

model_name: str

The model name in string. The default value is "resnet34". Refer to Timm Docs to get a full list of supported models.

num_classes: int

The number of classes. The default value is 1000. It is related to model and dataset.

skip_preprocess: bool

The flag to control whether to skip image pre-process. The default value is False. If set to True, it will skip image preprocessing steps (transforms). In this case, input image data must be prepared in advance in order to properly fit the model.

Interface

An image embedding operator takes a towhee image as input. It uses the pre-trained model specified by model name to generate an image embedding in ndarray.

Parameters:

data: towhee.types.Image

The decoded image data in towhee Image (a subset of numpy.ndarray).

Returns: numpy.ndarray

An image embedding generated by model, in shape of (feature_dim,).

save_model(format='pytorch', path='default')

Save model to local with specified format.

Parameters:

format: str

​ The format of saved model, defaults to 'pytorch'.

path: str

​ The path where model is saved to. By default, it will save model to the operator directory.

from towhee import ops

op = ops.image_embedding.timm(model_name='resnet50').get_op()
op.save_model('onnx', 'test.onnx')

supported_model_names(format=None)

Get a list of all supported model names or supported model names for specified model format.

Parameters:

format: str

​ The model format such as 'pytorch', 'torchscript'.

from towhee import ops


op = ops.image_embedding.timm().get_op()
full_list = op.supported_model_names()
onnx_list = op.supported_model_names(format='onnx')
print(f'Onnx-support/Total Models: {len(onnx_list)}/{len(full_list)}')

 2022-12-19 16:32:37,933 - 140704422594752 - timm_image.py-timm_image:88 - WARNING: The operator is initialized without specified model.
 Onnx-support/Total Models: 715/759

Fine-tune

To fine-tune this operator, please refer to this example guide.

Towhee Serve

Models which is supported the towhee.serve.

Model List

models	models	models
adv_inception_v3	bat_resnext26ts	beit_base_patch16_224
beit_base_patch16_224_in22k	beit_base_patch16_384	beit_large_patch16_224
beit_large_patch16_224_in22k	beit_large_patch16_384	beit_large_patch16_512
botnet26t_256	cait_m36_384	cait_m48_448
cait_s24_224	cait_s24_384	cait_s36_384
cait_xs24_384	cait_xxs24_224	cait_xxs24_384
cait_xxs36_224	cait_xxs36_384	coat_lite_mini
coat_lite_small	coat_lite_tiny	convit_base
convit_small	convit_tiny	convmixer_768_32
convmixer_1024_20_ks9_p14	convmixer_1536_20	convnext_base
convnext_base_384_in22ft1k	convnext_base_in22ft1k	convnext_base_in22k
convnext_large	convnext_large_384_in22ft1k	convnext_large_in22ft1k
convnext_large_in22k	convnext_small	convnext_small_384_in22ft1k
convnext_small_in22ft1k	convnext_small_in22k	convnext_tiny
convnext_tiny_384_in22ft1k	convnext_tiny_hnf	convnext_tiny_in22ft1k
convnext_tiny_in22k	convnext_xlarge_384_in22ft1k	convnext_xlarge_in22ft1k
convnext_xlarge_in22k	cs3darknet_focus_l	cs3darknet_focus_m
cs3darknet_l	cs3darknet_m	cspdarknet53
cspresnet50	cspresnext50	darknet53
deit3_base_patch16_224	deit3_base_patch16_224_in21ft1k	deit3_base_patch16_384
deit3_base_patch16_384_in21ft1k	deit3_huge_patch14_224	deit3_huge_patch14_224_in21ft1k
deit3_large_patch16_224	deit3_large_patch16_224_in21ft1k	deit3_large_patch16_384
deit3_large_patch16_384_in21ft1k	deit3_small_patch16_224	deit3_small_patch16_224_in21ft1k
deit3_small_patch16_384	deit3_small_patch16_384_in21ft1k	deit_base_distilled_patch16_224
deit_base_distilled_patch16_384	deit_base_patch16_224	deit_base_patch16_384
deit_small_distilled_patch16_224	deit_small_patch16_224	deit_tiny_distilled_patch16_224
deit_tiny_patch16_224	densenet121	densenet161
densenet169	densenet201	densenetblur121d
dla34	dla46_c	dla46x_c
dla60	dla60_res2net	dla60_res2next
dla60x	dla60x_c	dla102
dla102x	dla102x2	dla169
dm_nfnet_f0	dm_nfnet_f1	dm_nfnet_f2
dm_nfnet_f3	dm_nfnet_f4	dm_nfnet_f5
dm_nfnet_f6	dpn68	dpn68b
dpn92	dpn98	dpn107
dpn131	eca_botnext26ts_256	eca_halonext26ts
eca_nfnet_l0	eca_nfnet_l1	eca_nfnet_l2
eca_resnet33ts	eca_resnext26ts	ecaresnet26t
ecaresnet50t	ecaresnet269d	edgenext_small
edgenext_small_rw	edgenext_x_small	edgenext_xx_small
efficientnet_b0	efficientnet_b1	efficientnet_b2
efficientnet_b3	efficientnet_b4	efficientnet_el
efficientnet_el_pruned	efficientnet_em	efficientnet_es
efficientnet_es_pruned	efficientnet_lite0	efficientnetv2_rw_m
efficientnetv2_rw_s	efficientnetv2_rw_t	ens_adv_inception_resnet_v2
ese_vovnet19b_dw	ese_vovnet39b	fbnetc_100
fbnetv3_b	fbnetv3_d	fbnetv3_g
gc_efficientnetv2_rw_t	gcresnet33ts	gcresnet50t
gcresnext26ts	gcresnext50ts	gernet_l
gernet_m	gernet_s	ghostnet_100
gluon_inception_v3	gluon_resnet18_v1b	gluon_resnet34_v1b
gluon_resnet50_v1b	gluon_resnet50_v1c	gluon_resnet50_v1d
gluon_resnet50_v1s	gluon_resnet101_v1b	gluon_resnet101_v1c
gluon_resnet101_v1d	gluon_resnet101_v1s	gluon_resnet152_v1b
gluon_resnet152_v1c	gluon_resnet152_v1d	gluon_resnet152_v1s
gluon_resnext50_32x4d	gluon_resnext101_32x4d	gluon_resnext101_64x4d
gluon_senet154	gluon_seresnext50_32x4d	gluon_seresnext101_32x4d
gluon_seresnext101_64x4d	gluon_xception65	gmlp_s16_224
halo2botnet50ts_25
halo2botnet50ts_256	halonet26t	halonet50ts
haloregnetz_b	hardcorenas_a	hardcorenas_b
hardcorenas_c	hardcorenas_d	hardcorenas_e
hrnet_w18	hrnet_w18_small	hrnet_w18_small_v2
hrnet_w30	hrnet_w32	hrnet_w40
hrnet_w44	hrnet_w48	hrnet_w64
ig_resnext101_32x8d	ig_resnext101_32x16d	ig_resnext101_32x32d
ig_resnext101_32x48d	inception_resnet_v2	inception_v3
inception_v4	jx_nest_base	jx_nest_small
jx_nest_tiny	lambda_resnet26rpt_256	lambda_resnet26t
lambda_resnet50ts	lamhalobotnet50ts_256	lcnet_050
lcnet_075	lcnet_100	legacy_seresnet18
legacy_seresnet34	legacy_seresnet50	legacy_seresnet101
legacy_seresnet152	legacy_seresnext26_32x4d	levit_128
levit_128s	levit_192	levit_256
levit_384	mixer_b16_224	mixer_b16_224_in21k
mixer_b16_224_miil	mixer_b16_224_miil_in21k	mixer_l16_224
mixer_l16_224_in21k	mixnet_l	mixnet_m
mixnet_s	mixnet_xl	mnasnet_100
mnasnet_small	mobilenetv2_050	mobilenetv2_100
mobilenetv2_110d	mobilenetv2_120d	mobilenetv2_140
mobilenetv3_large_100	mobilenetv3_large_100_miil	mobilenetv3_large_100_miil_in21k
mobilenetv3_rw	mobilenetv3_small_050	mobilenetv3_small_075
mobilenetv3_small_100	mobilevit_s	mobilevit_xs
mobilevit_xxs	mobilevitv2_050	mobilevitv2_075
mobilevitv2_100	mobilevitv2_125	mobilevitv2_150
mobilevitv2_150_384_in22ft1k	mobilevitv2_150_in22ft1k	mobilevitv2_175
mobilevitv2_175_384_in22ft1k	mobilevitv2_175_in22ft1k	mobilevitv2_200
mobilevitv2_200_384_in22ft1k	mobilevitv2_200_in22ft1k	nf_regnet_b1
nf_resnet50	nfnet_l0	pit_b_224
pit_b_distilled_224	pit_s_224	pit_s_distilled_224
pit_ti_224	pit_ti_distilled_224	pit_xs_224
pit_xs_distilled_224	pnasnet5large	poolformer_m36
poolformer_m48	poolformer_s12	poolformer_s24
poolformer_s36	regnetv_040	regnetv_064
regnetx_002	regnetx_004	regnetx_006
regnetx_008	regnetx_016	regnetx_032
regnetx_040	regnetx_064	regnetx_080
regnetx_120	regnetx_160	regnetx_320
regnety_002	regnety_004	regnety_006
regnety_008	regnety_016	regnety_032
regnety_040	regnety_064	regnety_080
regnety_120	regnety_160	regnety_320
regnetz_040	regnetz_040h	regnetz_b16
regnetz_c16	regnetz_c16_evos	regnetz_d8
regnetz_d8_evos	regnetz_d32	regnetz_e8
repvgg_a2	repvgg_b0	repvgg_b1
repvgg_b1g4	repvgg_b2	repvgg_b2g4
repvgg_b3	repvgg_b3g4	res2net50_14w_8s
res2net50_26w_4s	res2net50_26w_6s	res2net50_26w_8s
res2net50_48w_2s	res2net101_26w_4s	res2next50
resmlp_12_224	resmlp_12_224_dino	resmlp_12_distilled_224
resmlp_24_224	resmlp_24_224_dino	resmlp_24_distilled_224
resmlp_36_224	resmlp_36_distilled_224	resmlp_big_24_224
resmlp_big_24_224_in22ft1k	resnest14d	resnest26d
resnest50d	resnest50d_1s4x24d	resnest50d_4s2x40d
resnest101e	resnest200e	resnet10t
resnet14t	resnet18	resnet18d
resnet26	resnet26d	resnet26t
resnet32ts	resnet33ts	resnet34
resnet34d	resnet50	resnet50_gn
resnet50d	resnet51q	resnet61q
resnet101	resnet101d	resnet152
resnet152d	resnet200d	resnetaa50
resnetblur50	resnetrs50

More Resources

• How to Get the Right Vector Embeddings - Zilliz blog: A comprehensive introduction to vector embeddings and how to generate them with popular open-source models.
• What Are Vector Embeddings?: Learn the definition of vector embeddings, how to create vector embeddings, and more.
• Embedding Inference at Scale for RAG Applications with Ray Data and Milvus - Zilliz blog: This blog showed how to use Ray Data and Milvus Bulk Import features to significantly speed up the vector generation and efficiently batch load them into a vector database.
• Image Embeddings for Enhanced Image Search - Zilliz blog: Image Embeddings are the core of modern computer vision algorithms. Understand their implementation and use cases and explore different image embedding models.
• Enhancing Information Retrieval with Sparse Embeddings | Zilliz Learn - Zilliz blog: Explore the inner workings, advantages, and practical applications of learned sparse embeddings with the Milvus vector database
• An Introduction to Vector Embeddings: What They Are and How to Use Them - Zilliz blog: In this blog post, we will understand the concept of vector embeddings and explore its applications, best practices, and tools for working with embeddings.