judgecliff SigLIP brand judge v1
Part of
judgecliff:
which image-QC judges survive optimization pressure?
A brand-fidelity judge for
rhode (beauty brand). A judge = these SigLIP weights
calibration.json (rhode train-split centroid + Platt scaling params) in this repo.
Base: google/siglip-so400m-patch14-384 (full SiglipModel, contrastively fine-tuned).
Training: SupCon contrastive fine-tune, rhode positives vs competitor-brand negatives (Glossier, ILIA and others), judgecliff train split. No violation negatives.
Report card (judgecliff Phase 1, 2,622-item test set): brand AUC 0.99, logo-masking delta 0.00 (style reader, not a wordmark reader), brand-dial Spearman 0.24, near-zero violation detection.
v1 learned the brand's center, not its boundaries: it is violation-blind, and under SRPO gradient pressure it was fully exploited (hacked images scored 0.84). See v2 (+violation negatives) and v3 (hardened) for the ablations.
Scoring
Score = Platt-calibrated cosine similarity between the image embedding and the rhode centroid:
1import json, numpy as np, torch
2from PIL import Image
3from transformers import SiglipModel, SiglipImageProcessor
4from huggingface_hub import hf_hub_download
5
6repo = "Gupta28/judgebench-siglip-judge-v1"
7model = SiglipModel.from_pretrained(repo)
8proc = SiglipImageProcessor.from_pretrained(repo)
9params = json.load(open(hf_hub_download(repo, "calibration.json")))
10centroid = np.array(params["centroid"]) # L2-normalized rhode train centroid
11
12img = Image.open("image.jpg")
13with torch.no_grad():
14 emb = model.get_image_features(**proc(images=img, return_tensors="pt"))
15v = emb[0].numpy(); v /= np.linalg.norm(v)
16cos = float(v @ centroid)
17score = 1 / (1 + np.exp(-(params["platt_a"] * cos + params["platt_b"]))) # calibrated on-brand prob
Full evaluation protocol, test-set construction, and findings:
https://github.com/amargupta0428/judgecliff.
