Search Results for "pixel"

...Because the objective is non-autoregressive and operates in embedding space, JEPA tends to be compute-efficient and stable at scale. The approach has become a strong alternative to contrastive or pixel-reconstruction methods for representation learning.

...Instead of reconstructing pixels, it predicts the missing high-level embeddings of masked space-time regions using a context encoder and a slowly updated target encoder. This objective encourages the model to learn semantics, motion, and long-range structure without the shortcuts that pixel-level losses can invite. The architecture is designed to scale: spatiotemporal ViT backbones, flexible masking schedules, and efficient sampling let it train on long clips while remaining stable. Trained representations transfer well to downstream tasks such as action recognition, temporal localization, and video retrieval, often with simple linear probes or light fine-tuning. ...

pytorch-fcn is a PyTorch implementation of Fully Convolutional Networks for semantic image segmentation. It reproduces the influential FCN approach that converts classification networks into dense, pixel-level predictors. The package includes FCN32s, FCN16s, FCN8s, and FCN8s-at-once variants with progressively finer output reconstruction. Training code and a PASCAL VOC example are provided so users can reproduce baseline experiments. The repository reports mean intersection-over-union results alongside the original implementation for comparison. ...