Unverified50% confidenceFactExact time
The modality gap is an empirical finding: after training, image embeddings and text embeddings cluster in two narrow cones on the unit sphere, with a stable angle between the two centroids.
1
Sources
50%
Confidence
Long-term
Relevance
10/26/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe embedding spaces of multimodal contrastive models (such as CLIP and CLAP) exhibit a modality gap: image and text embedding vectors each occupy their own region on the unit sphere and are difficult to align.68% similarUnverifiedThe convolution operation uses a small weight matrix (convolution kernel) to slide over the image, computing a weighted sum over a small region each time, and the output is called a feature map; different convolution kernels can detect different patterns.65% similarVerifiedCLIP is capable of semantically composing different visual features in vector space through compositional generalization, even when the exact description never appeared in the training data.63% similarUnverifiedIn a well-trained font embedding, distances in the vector space correspond to the visual similarity of fonts.63% similarUnverifiedDue to nonlinear radial projection, fisheye images accumulate a large number of redundant pixels at the image edges, causing encoding to require more bitrate, whereas half equirectangular projection has a more uniform pixel spatial distribution and better compatibility with HEVC block partitioning structures.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1098274API
curl https://kongchang.com/api/v1/knowledge/claims/1098274MCP
get_claim(id=1098274)