Brief-details: An anime-style LoRA model trained for generating Ozen character (Made in Abyss) images, featuring two versions - soft and hard, with "ozensama" trigger token
Brief-details: A YOLOv5-based object detection model specialized in hard hat detection, achieving 92.8% mAP@0.5, with easy deployment via PyTorch and comprehensive documentation.
Brief-details: BLEURT-20 is a PyTorch-based text classification model for evaluating text similarity with high accuracy, reaching scores of up to 0.999 in comparisons
BRIEF DETAILS: Japanese NER model based on LUKE, fine-tuned on Wikipedia data. 279M parameters, achieves 0.84 F1-score for entity recognition across 8 categories.
BRIEF-DETAILS: Russian image captioning model combining ViT encoder and ruGPT2 decoder, trained on translated COCO2014 dataset. First of its kind for Russian language.
Brief-details: A Russian-optimized Whisper ASR model fine-tuned on Common Voice 11.0, achieving 41.22% WER. Features 72.6M parameters and uses F32 tensors.
Brief-details: PPO-based reinforcement learning model for Crazyflie drone control in NVIDIA's Isaac Gym, achieving mean rewards of 1106.75 ±63.75 with optimized hyperparameters.
Brief-details: AV-HuBERT is a multimodal speech recognition model combining audio and visual inputs to enhance speech understanding through lip movement analysis.
Brief Details: A specialized NER model built on PubMedBERT for identifying gene and protein entities in biomedical text, trained on 19+ datasets with Apache 2.0 license.
Brief Details: A specialized Stable Diffusion model fine-tuned for generating Starcraft terrain maps, supporting 8 different tilesets with 64x64 resolution terrain generation.
Brief Details: German BERT large model optimized for sentence similarity, using euclidean distance metrics. 1024-dimensional embeddings, MIT licensed, ideal for few-shot classification.
Brief Details: Flair-based Persian Named Entity Recognition model achieving 90.33% F1-score on NSURL-2019, supporting 7 entity types including person, location, and organization names.
Brief Details: Powerful French ASR model fine-tuned from Whisper Large V2, achieving WER 4.03-8.66% across major benchmarks, trained on 2200+ hours of French speech.
BRIEF DETAILS: A semantic segmentation model based on SegFormer architecture, specifically trained for facade analysis with 12 distinct building element classes.
Brief-details: Japanese sentence embedding model based on LUKE, trained on JSNLI dataset. Generates 768-dimensional vectors for semantic search and clustering. Apache 2.0 licensed.
Brief Details: TunesFormer - Transformer-based dual-decoder model for Irish tune generation with control codes. Built for melody composition with form control.
Brief-details: A Stable Diffusion textual inversion model trained to generate Adolf Hitler-style imagery, built with MIT license and available through the SD-concepts-library.
Brief-details: A specialized Japanese-focused model conversion utility that transforms various AI models to fp16 format with safetensors, optimizing storage and compatibility
Brief-Details: One-shot talking face generation model that creates realistic facial animations from single reference image and audio, based on audio-visual correlation learning. Published at AAAI 2022.
Brief-details: UIE-base-en is an English information extraction model by PaddlePaddle, using ERNIE 3.0 for universal IE tasks including entity, relation, event, and sentiment extraction.
BRIEF DETAILS: Advanced semantic segmentation model with 216M parameters using Swin Transformer backbone, optimized for ADE20k dataset with masked attention mechanism
