OpenAI: CLIP

SigLIP is a vision-language model released in March 2023 by researchers at Google DeepMind. It adapts the CLIP image-text pretraining approach by replacing CLIP's softmax-based contrastive loss with a pairwise sigmoid loss, which operates independently on each image-text pair rather than requiring a global view of all pairs in a batch. This change decouples the loss from batch size, enabling more memory-efficient training and improved performance at smaller batch sizes, a regime where softmax contrastive learning typically struggles. Despite this simplification, SigLIP matches or exceeds CLIP-style models on zero-shot image classification and image-text retrieval benchmarks when trained on comparable data.SigLIP is distributed as an image encoder plus aligned text encoder, supporting zero-shot classification with arbitrary class vocabularies, image-text retrieval, and use as a frozen backbone in downstream vision-language models. Pretrained models are available at multiple Vision Transformer sizes and input resolutions, including 224, 256, 384, and 512 pixel inputs. SigLIP is released under the Apache 2.0 license by Google and is used as the vision encoder in Google's PaliGemma and PaliGemma 2. A successor, SigLIP 2, was released in February 2025 with multilingual support across 109 languages, improvements to localization and dense prediction, and two resolution handling variants (FixRes for backward-compatible fixed resolutions and NaFlex for native aspect ratio with variable sequence length).

DINOv2 is a self-supervised vision foundation model released in April 2023 by Meta AI's FAIR lab. It produces general-purpose visual features that transfer to a wide range of downstream tasks (including image classification, semantic segmentation, depth estimation, and image retrieval) without requiring task-specific fine-tuning. DINOv2 is trained on a curated dataset of 142 million images using a self-supervised objective combining student-teacher distillation, masked image modeling, and an image-level contrastive loss, extending the approach introduced in the original DINO.The model family spans Vision Transformer sizes from ViT-S (21M parameters) to ViT-g (1.1B parameters), with the larger variants setting state-of-the-art results on linear-probing benchmarks for classification, segmentation, and dense prediction tasks at release. DINOv2 features can be used directly as frozen backbones, reducing the need for labeled training data in downstream applications. The model is primarily used as an image encoder rather than as a complete task-specific model, making it a common backbone choice for custom vision pipelines. DINOv2 code and pretrained weights are released under the Apache 2.0 license, which was adopted after an initial CC-BY-NC 4.0 release in response to community requests for commercial compatibility. A successor model, DINOv3, was released in August 2025 with further scaling and a new training technique called Gram anchoring.

Vision Transformer is an image classification model developed by Google Research, first published in October 2020. It applies the transformer architecture directly to sequences of image patches without convolutional layers. Each image is divided into fixed-size patches, linearly projected into embeddings, and processed by a standard transformer encoder with multi-head self-attention. A classification token prepended to the patch sequence aggregates global image information for the final prediction.When pre-trained on large datasets such as JFT-300M and fine-tuned on ImageNet, ViT achieves competitive accuracy with state-of-the-art CNNs of the period. It performs best when pre-training data is abundant, as the lack of convolutional inductive biases makes it less data-efficient than CNN-based classifiers on smaller datasets. ViT established the foundation for transformer-based vision architectures and has influenced a broad range of subsequent models.

SAM-CLIP is a unified vision foundation model introduced by researchers at Apple and the University of Illinois Urbana-Champaign in October 2023. It merges two popular vision foundation models — Meta's Segment Anything Model (SAM) and OpenAI's CLIP — into a single shared Vision Transformer backbone through a combination of multi-task learning, continual learning, and teacher-student distillation. The method requires only a small fraction of the original pretraining datasets and demonstrates that complementary capabilities from distinct foundation models can be consolidated without retraining from scratch, reducing the storage and compute cost of running both models in inference.The resulting model retains SAM's zero-shot segmentation ability and CLIP's zero-shot classification and image-text retrieval, while introducing new capabilities the individual models lacked. SAM-CLIP establishes state-of-the-art results on zero-shot semantic segmentation across five benchmarks, improving mean IoU by 6.8 points on Pascal VOC and 5.9 points on COCO-Stuff over prior specialized models. The paper was accepted at the UniReps Workshop at NeurIPS 2023 and the eLVM Workshop at CVPR 2024. Apple has published the research but has not released model weights or inference code publicly.

ResNet-50 is a deep convolutional neural network architecture introduced in the 2015 paper "Deep Residual Learning for Image Recognition" by Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun at Microsoft Research. It is part of the ResNet (Residual Network) family, which introduced residual connections — shortcut paths that allow gradients to bypass layers during training — solving the degradation problem that had previously limited the practical training of very deep networks. ResNet-50 specifically refers to a 50-layer variant with approximately 25.6 million parameters, structured as a sequence of bottleneck residual blocks consisting of 1×1, 3×3, and 1×1 convolutions.ResNet-50 was trained on the ImageNet classification benchmark and achieved leading top-1 accuracy at release. Beyond classification, it became a widely used backbone feature extractor for downstream tasks including object detection (as the base network in Faster R-CNN, Mask R-CNN, and RetinaNet) and semantic and instance segmentation. Most current implementations in PyTorch torchvision, TensorFlow, and NVIDIA NGC use the ResNet-50 v1.5 variant, which relocates the stride-2 downsampling from the first 1×1 convolution to the 3×3 convolution within each bottleneck block, yielding approximately 0.5% higher top-1 accuracy than the original v1 formulation at a small throughput cost. ResNet-50 remains a common reference architecture in computer vision benchmarks and a standard backbone choice in detection and segmentation frameworks. The original Microsoft Research code is released under the MIT license.