LAXMAYDAY/pdm-pae-dinov2l-d32-imagenet256-train-full
收藏资源简介:
该数据集是一个用于PDM项目的PAE潜在缓存,包含从ImageNet-1k训练集(经过256x256 ADM风格裁剪)提取的DINOv2-L d32潜在表示。缓存包含1,281,167个样本,每个样本的潜在张量形状为[32, 16, 16],总大小为41.992 GB,以safetensors格式存储,分为313个分片。数据文件包含标签、潜在表示及其翻转版本。这是一个从ImageNet-1k训练图像派生的潜在缓存,使用时需遵守上游ImageNet条款及相关模型许可。
This dataset is a PAE latent cache for the PDM project, containing DINOv2-L d32 latent representations extracted from the ImageNet-1k training set (cropped in 256×256 ADM-style). The cache holds 1,281,167 samples, each with a latent tensor shape of [32, 16, 16], with a total size of 41.992 GB. It is stored in safetensors format and split into 313 shards. The data files include labels, latent representations, and their flipped versions. This is a latent cache derived from ImageNet-1k training images, and its use must comply with the upstream ImageNet terms and relevant model licenses.




