菜单
GPU-accelerated k-NN regression on CLIP image embeddings using cosine distance and inverse-distance-power weighting to predict target embedding vectors
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Memory-efficient Keras generator that streams sharded CSV files in chunks, renders strokes to images on-the-fly, and yields batches for training on datasets too large for memory
Stack 1D convolutions for local feature extraction before bidirectional LSTMs to classify variable-length stroke sequences into hundreds of doodle categories
Partition a large dataset into N balanced shards using integer key modulo arithmetic for reproducible, class-interleaved splits across CSV files
Normalize raw stroke coordinates to 0-255 range, resample at uniform arc-length spacing, then apply Ramer-Douglas-Peucker simplification
Render stroke sequences to grayscale images with temporal intensity encoding where earlier strokes are brighter and later strokes fade to encode drawing order
Score each face in a video as anomalous by computing L2 distance from the embedding centroid of all faces, then convert to probability via logistic function
| name | cv-embedding-knn-regression |
| description | GPU-accelerated k-NN regression on CLIP image embeddings using cosine distance and inverse-distance-power weighting to predict target embedding vectors |
When mapping images to target embeddings (e.g., prompt embeddings), k-NN regression in the CLIP embedding space is a strong non-parametric baseline. Compute cosine distances between test image embeddings and a reference set, find the k nearest neighbors, then weight their target embeddings by inverse distance raised to a power. GPU acceleration with batched matrix multiply makes this fast even for 100k+ reference points.
import torch
import numpy as np
def knn_predict(ref_x, ref_y, test_x, k=100, power=6, batch=1000):
ref_x = torch.from_numpy(ref_x).cuda()
ref_x /= ref_x.norm(dim=-1, keepdim=True)
ref_y = torch.from_numpy(ref_y).cuda()
preds = []
for i in range(0, len(test_x), batch):
batch_x = torch.from_numpy(test_x[i:i+batch]).cuda()
batch_x /= batch_x.norm(dim=-1, keepdim=True)
dists = 1 - batch_x @ ref_x.T # cosine distance
topk_dists, topk_idx = dists.topk(k, largest=False)
weights = 1.0 / (topk_dists ** power + 1e-8)
weights /= weights.sum(dim=-1, keepdim=True)
pred = (weights.unsqueeze(-1) * ref_y[topk_idx]).sum(dim=1)
preds.append(pred.cpu().numpy())
return np.concatenate(preds)