Unverified50% confidenceFactExact time
Token pruning refers to directly discarding some visual tokens based on importance scores, with a high compression rate but irreversibility; token merging instead aggregates similar tokens into representative vectors, preserving information while reducing sequence length.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedToken pruning methods discard key spatial cues, causing visual localization capability to be impaired.69% similarUnverifiedThe principle of tiled processing is to split the feature map into sub-tiles along the spatial dimension for independent computation and then merge them; peak VRAM usage decreases linearly as the number of tiles increases, but overlapping sampling is required to avoid seams and the total computation increases slightly.67% similarUnverifiedNeural Texture Compression encodes material textures as implicit representations in neural networks, preserving more detail while reducing video memory usage.67% similarUnverifiedCLIP compresses an entire image into a relatively low-dimensional semantic vector or a limited number of patch tokens, which may result in information loss regarding spatial layout and local details.67% similarUnverifiedCLIP编码器在跨模态对齐上存在细粒度语义丢失的局限，更擅长捕捉宏观风格特征而非精确空间位置关系65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1026736API
curl https://kongchang.com/api/v1/knowledge/claims/1026736MCP
get_claim(id=1026736)