Team Name
Big dataset of Cats
12.5k
Another dataset
12.5k
Biggest dataset
12.1m
Giant Dog Dataset
154k
Big dataset of Cats
12.5k
Big dataset of Cats
12.5k
Big dataset of Cats
12.5k
Big dataset of Cats
12.5k
Big dataset of Cats
12.5k
Big dataset of Cats
12.5k
Big dataset of Cats
12.5k
Big dataset of Cats
12.5k

New Model Editable Name

Neural Networks >

New Model Editable Name

Dataset

Dataset Name
Dataset Name

Version

Export Version Name
12/12/2019
Complete
1292 Images
12 Classes (no filters)
12
12
Export Version Name that takes longer
12/12/2019
Complete
1292 Images
12 Classes (no filters)
12
12
Short name
12/12/2019
Complete
1292 Images
12 Classes (no filters)
12
12
+ NEW VERSION

Model Type

Class name
882
Class name
882
Class name
882
Class name
882
+12 others

Instance Segmentation

Detects and segments objects that are annotated with either polygons or masks. Like an object detector, it can find individual instances of an object, which is useful for counting. Like a segmentation network, it defines their contour.

Characteristics of instance segmentation models:

  • Computationally Expensive
  • Good at counting
  • Good with many classes
  • Bad with continuous objects (sky, road, grass)
  • Good with discrete objects (person, cell, fracture)
  • Slower
Class name
882
Class name
882
Class name
882

Semantic segmentation

Paints in pixels belonging to a certain object annotated with either polygons or masks. For example: the floor, scratches, imperfections, water, and non-discrete entities. This neural network cannot count objects of the same class like in Instance Segmentation, but is significantly faster.

Characteristics of semantic segmentation models:

  • Computationally inexpensive
  • Good with few classes
  • Good with continuous objects (sky, road, grass)
  • Works on discrete objects (person, cell, fracture) but cannot distinguish instances.
  • Fast
Class name
882
Class name
882
Class name
882
Class name
882

Object Detection

Detects objects that are annotated with either polygons, masks, or bounding boxes. This network will output a bounding box around each object it finds. It will not segment them, but is faster than an instance segmentation network.

Characteristics of object detection models:

  • Medium computational expense
  • Good at counting
  • Good with many classes
  • Bad with continuous objects (sky, road, grass)
  • Good with discrete objects (person, cell, fracture)
  • Fast
Class name
882
Class name
882
Class name
882
Class name
882
+12 others

Image Classification

Predicts an image tag. Classification networks are slim, light, and quick to annotate for. Whilst they won't count objects or detect them in space, they become reliable with less training data and support many classes (tags).

Characteristics of classification models:

  • Computationally inexpensive
  • Good with many classes
  • Good with generic or abstract image characteristics (daytime, x-ray, blurry, source 21)
  • Outputs one tag, not multiple object labels.
  • Does not detect objects in space, only that they are likely to be present.
  • Very fast

Tier

Evaluation Models

A quick, small neural network that gives a glimpse of your dataset’s viability. These models train on a small number of images and have an unrefined performance for the purpose of feasibility testing.

Standard Models

These train on your full dataset for a large number of epochs. Use these during product development or to deploy in low to medium risk applications.

Production Models

One or more V7 deep learning engineers will be personally appointed to the training of this model. Select this option for high risk, custom, or unordinary applications where additional expertise is required.

Breakdown

Dataset Name
Version
Model Type
Tier
Estimated time to results
Up to

72 Hours

Total Cost

FREE

Estimated time to results
Up to

72 Hours

Total Cost

$999

Estimated time to results
Up to

72 Hours

Total Cost

Contact Us

TRAIN