Sketch-126-DomainNet

Sketch-126-DomainNet is an image classification vision-language encoder model fine-tuned from google/siglip2-base-patch16-224 for a single-label classification task. It is designed to classify sketches into 126 domain categories using the SiglipForImageClassification architecture.

👁 Sketch-126-DomainNet - visual selection.png

Moment Matching for Multi-Source Domain Adaptation : https://arxiv.org/pdf/1812.01754

SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features https://arxiv.org/pdf/2502.14786

Classification Report:
 precision recall f1-score support

 aircraft_carrier 1.0000 0.2200 0.3607 50
 alarm_clock 0.9873 0.9568 0.9718 162
 ant 0.9432 0.9326 0.9379 89
 anvil 0.2727 0.0423 0.0732 71
 asparagus 0.9673 0.8916 0.9279 166
 axe 0.8034 0.8773 0.8387 163
 banana 0.9744 0.9383 0.9560 162
 basket 0.7160 0.7682 0.7412 151
 bathtub 0.8073 0.9281 0.8635 167
 bear 0.8636 0.6690 0.7540 142
 bee 0.9196 0.8957 0.9075 115
 bird 0.9094 0.9429 0.9259 245
 blackberry 1.0000 0.1250 0.2222 48
 blueberry 0.6744 0.8529 0.7532 102
 bottlecap 0.7468 0.5315 0.6211 111
 broccoli 0.7727 0.9444 0.8500 144
 bus 0.9302 0.8989 0.9143 178
 butterfly 0.9594 0.9497 0.9545 199
 cactus 1.0000 0.6735 0.8049 49
 cake 0.0000 0.0000 0.0000 54
 calculator 0.9298 0.9636 0.9464 55
 camel 0.9208 0.8942 0.9073 104
 camera 0.9200 0.7931 0.8519 87
 candle 0.9556 0.6935 0.8037 62
 cannon 0.7500 0.2027 0.3191 74
 canoe 0.8000 0.5825 0.6742 103
 carrot 0.0000 0.0000 0.0000 27
 castle 0.9583 0.5111 0.6667 45
 cat 0.8961 0.6635 0.7624 104
 ceiling_fan 0.0000 0.0000 0.0000 20
 cell_phone 0.0000 0.0000 0.0000 18
 cello 0.9600 0.4706 0.6316 51
 chair 0.8043 0.4805 0.6016 77
 chandelier 0.0000 0.0000 0.0000 27
 coffee_cup 0.0000 0.0000 0.0000 26
 compass 0.0000 0.0000 0.0000 10
 computer 0.2500 0.0435 0.0741 23
 cow 0.0000 0.0000 0.0000 14
 crab 0.9123 0.8525 0.8814 122
 crocodile 0.9280 0.8992 0.9134 129
 cruise_ship 0.7467 0.9032 0.8175 124
 dog 0.8533 0.8911 0.8718 248
 dolphin 0.9091 0.8824 0.8955 68
 dragon 0.7914 0.8269 0.8088 156
 drums 0.9259 0.8772 0.9009 171
 duck 0.8409 0.8409 0.8409 220
 dumbbell 0.9507 0.9184 0.9343 147
 elephant 0.9630 0.9765 0.9697 213
 eyeglasses 0.8155 0.7919 0.8035 173
 feather 0.9344 0.9344 0.9344 244
 fence 0.8796 0.8482 0.8636 112
 fish 0.9527 0.9495 0.9511 297
 flamingo 0.9818 0.9474 0.9643 114
 flower 0.8267 0.9219 0.8717 269
 foot 0.7743 0.8578 0.8140 204
 fork 0.9366 0.9433 0.9399 141
 frog 0.9620 0.9383 0.9500 162
 giraffe 0.9655 0.9396 0.9524 149
 goatee 0.7914 0.8897 0.8377 145
 grapes 0.9132 0.9609 0.9364 230
 guitar 0.8462 0.9862 0.9108 145
 hammer 0.8333 0.4386 0.5747 57
 helicopter 0.9441 0.9620 0.9530 158
 helmet 0.8509 0.8204 0.8354 167
 horse 0.9091 0.9877 0.9467 81
 kangaroo 0.9592 0.9691 0.9641 97
 lantern 0.0000 0.0000 0.0000 30
 laptop 0.8273 0.9200 0.8712 250
 leaf 0.8449 0.8870 0.8655 301
 lion 0.9697 0.9734 0.9715 263
 lipstick 0.9634 0.8977 0.9294 88
 lobster 0.9265 0.9130 0.9197 138
 microphone 0.8917 0.8770 0.8843 122
 monkey 0.9297 0.8947 0.9119 133
 mosquito 0.9052 0.9211 0.9130 114
 mouse 0.8632 0.8039 0.8325 102
 mug 0.6928 0.7737 0.7310 137
 mushroom 0.8174 0.8861 0.8504 202
 onion 0.9538 0.9841 0.9688 126
 panda 0.9643 0.8710 0.9153 62
 peanut 0.8302 0.8462 0.8381 104
 pear 0.7966 0.9658 0.8731 146
 peas 0.6667 0.8438 0.7448 64
 pencil 0.0000 0.0000 0.0000 21
 penguin 0.9586 0.9701 0.9643 167
 pig 0.8983 0.8785 0.8883 181
 pillow 0.9570 0.9674 0.9622 92
 pineapple 0.9808 0.9714 0.9761 105
 potato 0.9444 0.5231 0.6733 65
 power_outlet 0.5556 0.0676 0.1205 74
 purse 0.9220 0.7182 0.8075 181
 rabbit 0.9697 0.8767 0.9209 73
 raccoon 0.7850 0.9097 0.8428 277
 rhinoceros 0.9863 0.9863 0.9863 146
 rifle 0.9143 0.9796 0.9458 98
 saxophone 0.9381 0.8618 0.8983 246
 screwdriver 0.7709 0.8706 0.8177 286
 sea_turtle 0.9698 0.9507 0.9602 203
 see_saw 0.3296 0.5738 0.4187 413
 sheep 0.9254 0.9153 0.9203 366
 shoe 0.9395 0.9688 0.9539 513
 skateboard 0.7365 0.7831 0.7591 332
 snake 0.8005 0.8737 0.8355 372
 speedboat 0.8388 0.8833 0.8605 377
 spider 0.7954 0.8696 0.8309 514
 squirrel 0.8511 0.8484 0.8498 310
 strawberry 0.8313 0.8471 0.8391 157
 streetlight 0.7944 0.8134 0.8038 209
 string_bean 0.7143 0.3000 0.4225 50
 submarine 0.5916 0.6975 0.6402 162
 swan 0.8966 0.8387 0.8667 186
 table 0.6705 0.7522 0.7090 230
 teapot 0.8464 0.8968 0.8709 252
 teddy-bear 0.6818 0.8385 0.7521 161
 television 0.8974 0.7071 0.7910 99
 the_Eiffel_Tower 0.9860 0.9679 0.9769 218
the_Great_Wall_of_China 0.6389 0.8440 0.7273 109
 tiger 0.9417 0.9604 0.9510 303
 toe 0.0000 0.0000 0.0000 53
 train 0.8650 0.9010 0.8827 192
 truck 0.8136 0.9372 0.8710 191
 umbrella 0.8650 0.8913 0.8779 230
 vase 0.8082 0.8082 0.8082 146
 watermelon 0.8947 0.8333 0.8629 102
 whale 0.8910 0.8744 0.8826 215
 zebra 0.9817 0.9727 0.9772 220

 accuracy 0.8440 19317
 macro avg 0.7818 0.7419 0.7475 19317
 weighted avg 0.8404 0.8440 0.8352 19317

The model categorizes images into the following 126 classes:

Class 0: "aircraft_carrier"
Class 1: "alarm_clock"
Class 2: "ant"
Class 3: "anvil"
Class 4: "asparagus"
Class 5: "axe"
Class 6: "banana"
Class 7: "basket"
Class 8: "bathtub"
Class 9: "bear"
Class 10: "bee"
Class 11: "bird"
Class 12: "blackberry"
Class 13: "blueberry"
Class 14: "bottlecap"
Class 15: "broccoli"
Class 16: "bus"
Class 17: "butterfly"
Class 18: "cactus"
Class 19: "cake"
Class 20: "calculator"
Class 21: "camel"
Class 22: "camera"
Class 23: "candle"
Class 24: "cannon"
Class 25: "canoe"
Class 26: "carrot"
Class 27: "castle"
Class 28: "cat"
Class 29: "ceiling_fan"
Class 30: "cell_phone"
Class 31: "cello"
Class 32: "chair"
Class 33: "chandelier"
Class 34: "coffee_cup"
Class 35: "compass"
Class 36: "computer"
Class 37: "cow"
Class 38: "crab"
Class 39: "crocodile"
Class 40: "cruise_ship"
Class 41: "dog"
Class 42: "dolphin"
Class 43: "dragon"
Class 44: "drums"
Class 45: "duck"
Class 46: "dumbbell"
Class 47: "elephant"
Class 48: "eyeglasses"
Class 49: "feather"
Class 50: "fence"
Class 51: "fish"
Class 52: "flamingo"
Class 53: "flower"
Class 54: "foot"
Class 55: "fork"
Class 56: "frog"
Class 57: "giraffe"
Class 58: "goatee"
Class 59: "grapes"
Class 60: "guitar"
Class 61: "hammer"
Class 62: "helicopter"
Class 63: "helmet"
Class 64: "horse"
Class 65: "kangaroo"
Class 66: "lantern"
Class 67: "laptop"
Class 68: "leaf"
Class 69: "lion"
Class 70: "lipstick"
Class 71: "lobster"
Class 72: "microphone"
Class 73: "monkey"
Class 74: "mosquito"
Class 75: "mouse"
Class 76: "mug"
Class 77: "mushroom"
Class 78: "onion"
Class 79: "panda"
Class 80: "peanut"
Class 81: "pear"
Class 82: "peas"
Class 83: "pencil"
Class 84: "penguin"
Class 85: "pig"
Class 86: "pillow"
Class 87: "pineapple"
Class 88: "potato"
Class 89: "power_outlet"
Class 90: "purse"
Class 91: "rabbit"
Class 92: "raccoon"
Class 93: "rhinoceros"
Class 94: "rifle"
Class 95: "saxophone"
Class 96: "screwdriver"
Class 97: "sea_turtle"
Class 98: "see_saw"
Class 99: "sheep"
Class 100: "shoe"
Class 101: "skateboard"
Class 102: "snake"
Class 103: "speedboat"
Class 104: "spider"
Class 105: "squirrel"
Class 106: "strawberry"
Class 107: "streetlight"
Class 108: "string_bean"
Class 109: "submarine"
Class 110: "swan"
Class 111: "table"
Class 112: "teapot"
Class 113: "teddy-bear"
Class 114: "television"
Class 115: "the_Eiffel_Tower"
Class 116: "the_Great_Wall_of_China"
Class 117: "tiger"
Class 118: "toe"
Class 119: "train"
Class 120: "truck"
Class 121: "umbrella"
Class 122: "vase"
Class 123: "watermelon"
Class 124: "whale"
Class 125: "zebra"

Run with Transformers🤗

!pip install -q transformers torch pillow gradio

import gradio as gr
from transformers import AutoImageProcessor, SiglipForImageClassification
from transformers.image_utils import load_image
from PIL import Image
import torch

# Load model and processor
model_name = "prithivMLmods/Sketch-126-DomainNet"
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

def sketch_classification(image):
 """Predicts the sketch category for an input image."""
 # Convert the input numpy array to a PIL Image and ensure it has 3 channels (RGB)
 image = Image.fromarray(image).convert("RGB")
 
 # Process the image and prepare it for the model
 inputs = processor(images=image, return_tensors="pt")
 
 # Perform inference without gradient calculation
 with torch.no_grad():
 outputs = model(**inputs)
 logits = outputs.logits
 # Convert logits to probabilities using softmax
 probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist()
 
 # Mapping from indices to corresponding sketch category labels
 labels = {
 "0": "aircraft_carrier", "1": "alarm_clock", "2": "ant", "3": "anvil", "4": "asparagus",
 "5": "axe", "6": "banana", "7": "basket", "8": "bathtub", "9": "bear",
 "10": "bee", "11": "bird", "12": "blackberry", "13": "blueberry", "14": "bottlecap",
 "15": "broccoli", "16": "bus", "17": "butterfly", "18": "cactus", "19": "cake",
 "20": "calculator", "21": "camel", "22": "camera", "23": "candle", "24": "cannon",
 "25": "canoe", "26": "carrot", "27": "castle", "28": "cat", "29": "ceiling_fan",
 "30": "cell_phone", "31": "cello", "32": "chair", "33": "chandelier", "34": "coffee_cup",
 "35": "compass", "36": "computer", "37": "cow", "38": "crab", "39": "crocodile",
 "40": "cruise_ship", "41": "dog", "42": "dolphin", "43": "dragon", "44": "drums",
 "45": "duck", "46": "dumbbell", "47": "elephant", "48": "eyeglasses", "49": "feather",
 "50": "fence", "51": "fish", "52": "flamingo", "53": "flower", "54": "foot",
 "55": "fork", "56": "frog", "57": "giraffe", "58": "goatee", "59": "grapes",
 "60": "guitar", "61": "hammer", "62": "helicopter", "63": "helmet", "64": "horse",
 "65": "kangaroo", "66": "lantern", "67": "laptop", "68": "leaf", "69": "lion",
 "70": "lipstick", "71": "lobster", "72": "microphone", "73": "monkey", "74": "mosquito",
 "75": "mouse", "76": "mug", "77": "mushroom", "78": "onion", "79": "panda",
 "80": "peanut", "81": "pear", "82": "peas", "83": "pencil", "84": "penguin",
 "85": "pig", "86": "pillow", "87": "pineapple", "88": "potato", "89": "power_outlet",
 "90": "purse", "91": "rabbit", "92": "raccoon", "93": "rhinoceros", "94": "rifle",
 "95": "saxophone", "96": "screwdriver", "97": "sea_turtle", "98": "see_saw", "99": "sheep",
 "100": "shoe", "101": "skateboard", "102": "snake", "103": "speedboat", "104": "spider",
 "105": "squirrel", "106": "strawberry", "107": "streetlight", "108": "string_bean",
 "109": "submarine", "110": "swan", "111": "table", "112": "teapot", "113": "teddy-bear",
 "114": "television", "115": "the_Eiffel_Tower", "116": "the_Great_Wall_of_China",
 "117": "tiger", "118": "toe", "119": "train", "120": "truck", "121": "umbrella",
 "122": "vase", "123": "watermelon", "124": "whale", "125": "zebra"
 }
 
 # Create a dictionary mapping each label to its predicted probability (rounded)
 predictions = {labels[str(i)]: round(probs[i], 3) for i in range(len(probs))}
 return predictions

# Create Gradio interface
iface = gr.Interface(
 fn=sketch_classification,
 inputs=gr.Image(type="numpy"),
 outputs=gr.Label(label="Prediction Scores"),
 title="Sketch-126-DomainNet Classification",
 description="Upload a sketch to classify it into one of 126 categories."
)

# Launch the app
if __name__ == "__main__":
 iface.launch()

Intended Use:

The Sketch-126-DomainNet model is designed for sketch image classification. It is capable of categorizing sketches into a wide range of domains—from objects like an "aircraft_carrier" or "alarm_clock" to animals, plants, and everyday items. Potential use cases include:

Art and Design Applications: Assisting artists and designers in organizing and retrieving sketches based on content.
Creative Search Engines: Enabling sketch-based search for design inspiration.
Educational Tools: Helping students and educators in art and design fields with categorization and retrieval of visual resources.
Computer Vision Research: Providing a benchmark dataset for sketch recognition and domain adaptation tasks.