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

Lemniscate.pytorch

Lemniscate.pytorch is a Python and PyTorch implementation of non-parametric instance discrimination for unsupervised image feature learning. It trains ResNet-based encoders with negative samples and a memory bank, and provides pretrained model weights and image feature representations.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Lemniscate.pytorch — 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 research implementation for learning image features without labels, or want to evaluate pretrained representations with nearest-neighbor methods. The README provides separate ImageNet and CIFAR-10 training instructions.

Who it is for

  • Computer-vision researchers studying self-supervised or unsupervised image representations.
  • Practitioners training image encoders or evaluating learned features with nearest-neighbor methods.
  • Developers who want pretrained ResNet weights or existing ImageNet feature representations for experiments.

Documented uses

  • Train an image encoder on ImageNet or CIFAR-10 using the repository’s documented commands.
  • Adjust negative-sample count, temperature, momentum and embedding dimension in ImageNet training.
  • Evaluate a saved ImageNet model using nearest-neighbor classification.
  • Start experiments from pretrained ResNet weights or use supplied ImageNet feature representations.
  • Inspect the project’s ImageNet nearest-neighbor retrieval comparisons.
Before you choose

Requirements & limits

The implementation is available in a public GitHub repository with clone instructions.

  • The README specifies Python 2.7 and PyTorch 0.4; PyTorch 0.3 users are directed to a separate tag.
  • ImageNet data preparation and hardware configuration rely on the linked PyTorch ImageNet example.
  • The updated ResNet-50 checkpoint uses 65,536 negatives and Softmax cross-entropy, while older ResNet releases use 4,096 negatives and NCE loss.
  • The repository includes separate ImageNet and CIFAR-10 training commands; the documented ImageNet testing command defaults to 200 nearest neighbors. The

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

Train an instance-discrimination model on ImageNet

Clone the repository and start ImageNet training with a ResNet-18 encoder.

Before you start

  • Python 2.7 and PyTorch 0.4 are listed as supported.
  • ImageNet data prepared according to the referenced PyTorch ImageNet example; hardware guidance is also referenced there.
  • A local repository clone and the ImageNet data path (DATAPATH).

Steps

  1. Clone the repository: the command below.

    Documented command
    git clone https://github.com/zhirongw/lemniscate.pytorch

  2. Run the command below.

    Documented command
    python main.py DATAPATH --arch resnet18 -j 32 --nce-k 4096 --nce-t 0.07 --lr 0.03 --nce-m 0.5 --low-dim 128 -b 256

  3. During training, use the described K=1 nearest-neighbor validation monitoring as an estimate of feature quality.

Expected outcome

The documented training procedure uses 4,096 negative samples and a 128-dimensional embedding; training monitors feature quality with K=1 nearest-neighbor validation.

Guide limitations

  • The README refers data preparation and hardware configuration to the official PyTorch ImageNet example rather than detailing them here.
  • The listed environment is Python 2.7 and PyTorch 0.4; PyTorch 0.3 users are directed to tag v0.3.
  • The README documents the training command, not a verified run or its completion.
Documentation 1 ↗
Source-based practical guide

Test an ImageNet model with nearest neighbors

Run the documented ImageNet evaluation using a ResNet-18 model checkpoint.

Before you start

  • Python 2.7 and PyTorch 0.4 are listed as supported.
  • ImageNet data prepared according to the referenced PyTorch ImageNet example.
  • A model checkpoint path (input_model.pth.tar) and the ImageNet data path (DATAPATH).

Steps

  1. Clone the repository: the command below.

    Documented command
    git clone https://github.com/zhirongw/lemniscate.pytorch

  2. Run the command below.

    Documented command
    python main.py DATAPATH --arch resnet18 --resume input_model.pth.tar -e

  3. Review the evaluation using the documented default of K=200 neighbors.

Expected outcome

The documented command runs ImageNet testing with a default nearest-neighbor count of K=200.

Guide limitations

  • Data preparation and hardware configuration are referred to the official PyTorch ImageNet example.
  • The command is documented, but no successful execution is verified.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
18 October 2026
Original website title
zhirongw/lemniscate.pytorch · Official homepage title · retained in its source language
How we review sourcesSuggest a correction