voxel51/fiftyone

[FR] Add remaining models from TF2 object detection model zoo

Open

#746 opened on Jan 7, 2021

View on GitHub
 (5 comments) (0 reactions) (0 assignees)Python (400 forks)batch import
coreenhancementgood first issuemodel zoo

Repository metrics

Stars
 (4,021 stars)
PR merge metrics
 (Avg merge 3d 11h) (161 merged PRs in 30d)

Description

Currently, FiftyOne only contains CenterNet 512x512 with the Hourglass104 backbone from the TF2 model zoo: https://github.com/tensorflow/models/blob/archive/research/object_detection/g3doc/tf2_detection_zoo.md

Other models can be added by updating fiftyone/zoo/models/manifest-tf.json to add relevant information for the new models following centernet-hg104-512-coco-tf2.

It can then be tested with:

conda create -n fo
conda activate fo
conda install tensorflow-gpu=2.2.0
conda install ipython

git clone https://github.com/voxel51/fiftyone.git
cd fiftyone
bash install.bash
eta install models

# Add new models

import fiftyone as fo
import fiftyone.zoo as foz

dataset, session = fo.quickstart()

model = foz.load_zoo_model("NEW_MODEL_NAME", error_level=1)

dataset.apply_model(model, label_field="predictions")

session = fo.launch_app(dataset)

Contributor guide

Research direction
Look at the existing entry for 'centernet hg104-512 coco tf2' in the manifest tf.json file, replicate its structure for a new model from the TF2 object detection zoo, and then test by loading the model with FiftyOne.
Tech stack
pythontensorflow
Domain
backendcomputer vision
Issue type
Feature
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
2
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-3 hours
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
Active
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
Clear
Prerequisites
GitJSON
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
80