OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Kaggle Imaterialist

Kaggle Imaterialist is a public Python computer-vision repository documenting a winning iMaterialist Fashion 2019 instance-segmentation solution. It combines a Hybrid Task Cascade model with a ResNeXt-101-64x4d-FPN backbone, MMDetection-based training, test-time augmentation, checkpoint ensembling and mask postprocessing.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Kaggle Imaterialist — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use it when you need a concrete reference implementation for fashion-image instance segmentation, especially a pipeline that documents both training preparation and ensemble inference.

Who it is for

  • Computer-vision practitioners studying fashion-item instance segmentation.
  • Developers adapting competition code for image-dataset training and inference.
  • Readers examining multi-scale test-time augmentation, checkpoint ensembles and mask cleanup.

Documented uses

  • Prepare the competition data and annotations for MMDetection.
  • Train a fashion instance-segmentation model from COCO-pretrained weights.
  • Run test-time predictions at multiple scales with horizontal flips.
  • Combine predictions from three selected checkpoints and postprocess masks.
Before you choose

Requirements & limits

The repository is public and identifies its license as MIT.

  • The project is a code repository, not a hosted image-upload service.
  • The model and workflow depend on MMDetection, prepared dataset annotations, pretrained weights and GPU-oriented training scripts.
  • The README's recorded training configuration uses eight Tesla V100 GPUs; it does not establish a minimum hardware requirement.
  • The solution is tuned to the iMaterialist Fashion 2019 competition, including decisions about classes and postprocessing.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare the repository for model training

Set up the documented MMDetection build, weights, and dataset annotations before starting training.

Before you start

  • Docker workflow documented by the repository.
  • iMaterialist data arranged with train, test, and train.csv.zip under /data/.
  • Python and GPU training environment; the author's recorded setup used eight Tesla V100 GPUs.

Steps

  1. Build and enter the Docker environment using the repository's documented make build, make run-[server-name], and make exec commands.

  2. Build MMDetection by running compile.sh and python setup.py develop from its directory.

  3. Run prepare_weights.sh, then place the dataset in the documented /data/ structure and address the noted train.csv.zip error.

  4. Run the train and test annotation preparation scripts and split.sh from scripts, then use dist_train.sh with a configuration and GPU count to train.

Expected outcome

The documented preparation stages produce MMDetection annotations and lead to a distributed training command for the fashion segmentation model.

Guide limitations

  • The guide covers documented preparation and command structure; it does not verify successful execution or model output.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
amirassov/kaggle-imaterialist · Official homepage title · retained in its source language
How we review sourcesSuggest a correction