PG-COCO-ViT-b16

Pictograph Research · Classification · Trained 15m ago · Custom license

Use this model

PG-COCO-ViT-b16

Environment

Weights

ONNX SafeTensors PyTorch ExecuTorch TensorRT Engine

Precision

FP32 FP16

Model weights Model config

from pictograph import Client, get_model, ClassificationModel, ClassificationResult

client = Client(api_key="pk_live_...")   # Settings -> API Keys

# Downloads + caches the weights on first use
model: ClassificationModel = get_model(
    name="PG-COCO-ViT-b16",
    task="classification",
    format="onnx",
    client=client,
)
result: ClassificationResult = model.predict(image="photo.jpg")

print(result.top.name, round(result.top.confidence, 2))

from pictograph import load_model, ClassificationModel, ClassificationResult

# Download the weights + config with the buttons above
model: ClassificationModel = load_model(
    weights="/path/to/weights.onnx",
    config="/path/to/config.json",
    task="classification",
    format="onnx",
    device="auto",
)
result: ClassificationResult = model.predict(image="photo.jpg")

print(result.top.name, round(result.top.confidence, 2))

Deploy this model to Pictograph for an always-on, autoscaling /predict endpoint. Then call it from anywhere:

curl -X POST "https://workflows.pictograph.io/your-endpoint/predict" \
  -H "Authorization: Bearer pk_deploy_..." \
  -F "file=@frame.jpg"

from pictograph import DeploymentClient, ClassificationResult

client = DeploymentClient(
    endpoint="https://workflows.pictograph.io/your-endpoint/predict",
    api_key="pk_deploy_...",  # from your deployment's settings
    task="classification",
)
result: ClassificationResult = client.infer(image="frame.jpg")

print(result.top.name, round(result.top.confidence, 2))

pictograph deployments predict \
  <deployment-id> \
  frame.jpg \
  --token pk_deploy_... \
  --endpoint "https://workflows.pictograph.io/your-endpoint/predict"

Details

TaskClassification
Architecturevit_b_16
Precisionfp32
Version1.0.0
Trained15m ago
OrganizationPictograph Research

Evaluation

72.4%ACCURACY

62.2%F1

61.4%RECALL

66.2%PRECISION

Classes 55

skateboarding skiing bathroom bird frisbee giraffe soccer teddy bear elephant vintage photo bear zebra tennis train baseball cat dog airplane kitchen kitesurfing kite boat beach surfing horse vase livestock bedroom fruit living room market motorcycle portrait clock street fire hydrant bus luggage bench pizza office electronics child video games dining formal wear donut dessert meal vegetables cake hot dog street sign stop sign other

⚡ Inference

Runs free on shared CPU, right in the app.

Drop an image to try it

Confidence ≥ 0.50

Model Card

PG-COCO-ViT-b16

A classification model (vit_b_16), trained on 55 classes, scoring 72.4% Accuracy.

Pictograph Research · Classification · vit_b_16 · FP32 · a custom license · Public

Overview

This is a classification model built on vit_b_16, trained on the COCO-CLS-5k dataset. It expects 224x224 input and runs in FP32 precision. It predicts skateboarding, skiing, bathroom, bird, frisbee, giraffe, and 49 more.

Performance

Headline: 72.4% Accuracy.

Metric	Value
Accuracy	72.4%
F1	62.2%
Precision	66.2%
Recall	61.4%

Per-class Accuracy

Class	Accuracy
airplane	94.4%
baseball	85.2%
bathroom	80.0%
beach	0.0%
bear	66.7%
bedroom	40.0%
bench	42.9%
bird	88.9%
boat	78.6%
bus	74.1%
cake	0.0%
cat	90.9%
child	0.0%
clock	91.7%
dessert	0.0%
dining	59.1%
dog	81.2%
donut	50.0%
electronics	66.7%
elephant	100.0%
fire hydrant	70.0%
formal wear	61.5%
frisbee	69.2%
fruit	78.6%
giraffe	90.0%
horse	73.7%
hot dog	50.0%
kitchen	63.3%
kite	62.5%
kitesurfing	0.0%

Classes

Class index matches the model's output order.

#	Class
0	skateboarding
1	skiing
2	bathroom
3	bird
4	frisbee
5	giraffe
6	soccer
7	teddy bear
8	elephant
9	vintage photo
10	bear
11	zebra
12	tennis
13	train
14	baseball
15	cat
16	dog
17	airplane
18	kitchen
19	kitesurfing
20	kite
21	boat
22	beach
23	surfing
24	horse
25	vase
26	livestock
27	bedroom
28	fruit
29	living room
30	market
31	motorcycle
32	portrait
33	clock
34	street
35	fire hydrant
36	bus
37	luggage
38	bench
39	pizza
40	office
41	electronics
42	child
43	video games
44	dining
45	formal wear
46	donut
47	dessert
48	meal
49	vegetables
50	cake
51	hot dog
52	street sign
53	stop sign
54	other

Training

Setting	Value
Base dataset	COCO-CLS-5k
Export	COCO-CLS-5k-base
Dataset size	5,000 images
Input size	224x224
Epochs	40
Batch size	32
Precision	FP32
Version	1.0.0

Use this model

Run inference with the Pictograph SDK:

from pictograph import Client, get_model, ClassificationModel, ClassificationResult

client = Client(api_key="pk_live_...")  # Settings -> API Keys

# Downloads + caches the weights on first use
model: ClassificationModel = get_model(
    name="PG-COCO-ViT-b16",
    task="classification",
    format="onnx",
    precision="fp32",
    device="cpu",
    client=client,
)
result: ClassificationResult = model.predict(image="photo.jpg")
print(result)

# Fully offline (no API key) once you have the weights + config.json
from pictograph import load_model, ClassificationModel, ClassificationResult

model: ClassificationModel = load_model(
    "/path/to/weights.onnx",
    "/path/to/config.json",
    task="classification",
    format="onnx",
)
result: ClassificationResult = model.predict(image="photo.jpg")

Full inference guide: local inference docs.
REST + CLI: models API reference.
Deploy it as an always-on endpoint: deployments.

License

Released under a custom license.

Download the model files

Exports to ONNX. Sign in to download the weights and config, or pull it from the SDK with the Use Model snippet.

COCO-CLS-5k

Pictograph Research

5,000 images · 0 classes

Custom License Other

Non-commercial use only. Trained on COCO-CLS-5k, whose images are COCO's: the COCO Consortium does not own their copyright; each image comes from Flickr under its own license, recorded per image in COCO's annotation files, and most of them (68.9% of train2017) are Creative Commons NonCommercial. COCO's annotations are CC BY 4.0, the COCO Consortium. Cite Lin et al., Microsoft COCO: Common Objects in Context (ECCV 2014).

Join the discussion

Comments and reactions live in the app.

Build your first vision tool

$5 of free compute every month. No card required.

Start free View pricing