SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2310.05916 · 2023

Interpreting CLIP's Image Representation via Text-Based Decomposition

arXiv · PDF · Open in the Atlas

Code that ran

1 of 1 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 0 failures of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2310.05916.json · badge: /api/ran/2310.05916.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
yossigandelsman/clip_text_span canonical (classifier label)
classifier basis: classifier_2025
1 ran · 0 no recorded run
FunctionStatusWhere it lives
zero_shot_classifier Ran
ran, honoured the contract we drafted
yossigandelsman/clip_text_span/compute_text_projection.py
code served (permissive licence) · get_code("2901a70c780c9095")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (4, 3) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.nn import functional as F

torch.manual_seed(0)

# Build a minimal CLIP-like model with encode_text method
class MinimalCLIPModel(nn.Module):
    def __init__(self, embedding_dim=4):
        super().__init__()
        self.embedding_dim = embedding_dim
        self.text_encoder = nn.Linear(10, embedding_dim)
    
    def encode_text(self, text_tokens):
        # text_tokens shape: (batch_size, seq_len) of integers
        # We'll just use a simple embedding + linear layer
        # Create a simple embedding: treat tokens as features
        batch_size = text_tokens.shape[0]
        # Simple encoding: sum over sequence dimension
        encoded = self.text_encoder(text_tokens.float().mean(dim=1, keepdim=True).expand(-1, 10))
        return encoded

# Build a minimal tokenizer that converts strings to token tensors
class MinimalTokenizer:
    def __call__(self, texts):
        # Convert list of strings to token tensor
        # Simple approach: create random token sequences of fixed length
        batch_size = len(texts)
        seq_len = 8
        tokens = torch.randint(0, 100, (batch_size, seq_len))
        return tokens

# Set up parameters
device = torch.device('cpu')
model = MinimalCLIPModel(embedding_dim=4)
model.eval()

tokenizer = MinimalTokenizer()

classnames = ['dog', 'cat', 'bird']

# Templates can either use .format(c=classname) or be callable
templates = [
    'a photo of a {}',
    'a picture of a {}',
    'an image of a {}'
]

# Since use_format=False by default in the function signature,
# we need callable templates that work with template(classname)
templates = [
    lambda c, t='a photo of a {}': t.format(c),
    lambda c, t='a picture of a {}': t.format(c),
    lambda c, t='an image of a {}': t.format(c),
]

ARGS = {
    'model': model,
    'tokenizer': tokenizer,
    'classnames': classnames,
    'templates': templates,
    'device': device,
    'amp': True,
    'use_format': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (4, 3) — shape (4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 1 samples lifted from this paper's linked repositories, 1 carry a function identity and 0 do not. Of the 1 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 1 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We investigate the CLIP image encoder by analyzing how individual model components affect the final representation. We decompose the image representation as a sum across individual image patches, model layers, and attention heads, and use CLIP's text representation to interpret the summands. Interpreting the attention heads, we characterize each head's role by automatically finding text representations that span its output space, which reveals property-specific roles for many heads (e.g. location or shape). Next, interpreting the image patches, we uncover an emergent spatial localization within CLIP. Finally, we use this understanding to remove spurious features from CLIP and to create a strong zero-shot image segmenter. Our results indicate that a scalable understanding of transformer models is attainable and can be used to repair and improve models.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2310.05916")
get_code_for_paper("2310.05916")
have("2310.05916")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2310.05916.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2310.05916.svg)](https://syntology.ai/paper/2310.05916)

Connect an agent — have() is free.