SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2111.13792 · 2021

LAFITE: Towards Language-Free Training for Text-to-Image Generation

arXiv · PDF · Open in the Atlas

Code that ran

12 of 18 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; 2 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 2 failures of our instrument; 6 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/2111.13792.json · badge: /api/ran/2111.13792.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
drboog/Lafite canonical (classifier label)
classifier basis not recorded on the link
3 ran · 3 no recorded run
oxygenlu/ratlip community (classifier label)
classifier basis not recorded on the link
9 ran · 3 no recorded run
FunctionStatusWhere it lives
MA_GP_FP32 Ran
ran on a synthesized input
oxygenlu/ratlip/code/lib/modules.py
code served (permissive licence) · get_code("a2f5ae76d8f7b686")
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 0, dtype family float, shape () 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

torch.manual_seed(0)

# The function computes gradients w.r.t. img and sent, then computes a gradient penalty loss
# img and sent must be tensors with requires_grad=True
# out must be a scalar or tensor that can have gradients computed through it

# Create small tensors
batch_size = 2
img_dim = 4
sent_dim = 3

img = torch.randn(batch_size, img_dim, requires_grad=True)
sent = torch.randn(batch_size, sent_dim, requires_grad=True)

# Create an output tensor by combining img and sent through a simple operation
# The output should be differentiable w.r.t. both img and sent
out = torch.sum(img) + torch.sum(sent)

ARGS = {
    "img": img,
    "sent": sent,
    "out": out
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () 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
autopad Ran
ran, honoured the contract we drafted
oxygenlu/ratlip/code/models/RATLIP.py
pointer only (licence: MIT); the stored copy get_code serves, from yang-0201/YOLOv6_pro, is pointer only (licence: GPL-3.0) · get_code("988a3c854b1b13d0")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from yang-0201/YOLOv6_pro, and GPL-3.0, as GitHub detects it on that repository, is not in the permissive set Syntology serves inline. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

custom_reshape Ran
ran on a synthesized input
drboog/Lafite/dataset_tool.py
code served (permissive licence) · get_code("080da024bbcdf7bc")
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 4, dtype family float, shape (2, 3, 224, 224) as a later run recorded it. The image it ran in was 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.functional as F

torch.manual_seed(0)

# Create a small test image tensor
# Shape: (batch=2, channels=3, height=256, width=256)
img = torch.randn(2, 3, 256, 256)

ARGS = {
    'img': img,
    'mode': 'bicubic',
    'ratio': 0.99
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 3, 224, 224) — shape (2, 3, 224, 224) 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
file_ext Ran
ran; our contract draft was wrong, not the code
drboog/Lafite/dataset_tool.py
pointer only (licence: MIT); the stored copy get_code serves, from pdillis/stylegan3-fun, is pointer only (licence: NOASSERTION) · get_code("a2b45afa097b55b6")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from pdillis/stylegan3-fun, and GitHub could not classify the licence of that repository (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_D_in_out_chs Ran
ran on a synthesized input
oxygenlu/ratlip/code/models/RATLIP.py
code served (permissive licence) · get_code("f8cd9b9545621bc0")
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 zip. The image it ran in was not recorded. If your copy returns the same type 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 numpy as np

# The function takes two parameters:
# - nf: a number representing base number of filters/channels
# - imsize: an image size (typically power of 2)

# Looking at the function body:
# - nf is used in multiplication: nf * min(2 ** idx, 8)
# - imsize is used in np.log2(imsize) to compute layer_num
# The function returns pairs of (in_channels, out_channels)

# Valid values:
nf = 64  # typical base filter count in discriminators
imsize = 32  # power of 2, reasonable image size

ARGS = {
    "nf": nf,
    "imsize": imsize,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type zip — a zip output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a zip output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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
get_G_in_out_chs Ran
ran; our contract draft was wrong, not the code
oxygenlu/ratlip/code/models/RATLIP.py
code served (permissive licence) · get_code("7b1f9c201441fa1c")
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) and returned zip. The image it ran in was not recorded. If your copy returns the same type 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 numpy as np

ARGS = {
    "nf": 64,
    "imsize": 256,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type zip — a zip output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a zip output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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
get_fix_data Ran
ran on a synthesized input
oxygenlu/ratlip/code/lib/datasets.py
code served (permissive licence) · get_code("1b820b0f9416f8b5")
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 tuple. The image it ran in was not recorded. If your copy returns the same type 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
import torch.utils.data as data

torch.manual_seed(0)

# Define a minimal text encoder that implements the __call__ interface
class SimpleTextEncoder(nn.Module):
    def __init__(self, sent_dim=5, word_dim=6):
        super().__init__()
        self.sent_dim = sent_dim
        self.word_dim = word_dim
    
    def forward(self, caption):
        # caption is a tensor of shape [batch_size, seq_len]
        batch_size = caption.shape[0]
        sent_emb = torch.randn(batch_size, self.sent_dim)
        words_embs = torch.randn(batch_size, 10, self.word_dim)  # 10 words per caption
        return sent_emb, words_embs

# Create a simple dataset
class SimpleDataset(data.Dataset):
    def __init__(self, num_samples=4):
        self.num_samples = num_samples
    
    def __len__(self):
        return self.num_samples
    
    def __getitem__(self, idx):
        imgs = torch.randn(3, 32, 32)  # 3-channel image
        captions = f"caption_{idx}"
        CLIP_tokens = torch.randint(0, 100, (77,))  # typical CLIP token length
        keys = f"key_{idx}"
        return imgs, captions, CLIP_tokens, keys

# Create dataloaders
train_dataset = SimpleDataset(num_samples=2)
test_dataset = SimpleDataset(num_samples=2)

def collate_fn(batch):
    imgs = torch.stack([item[0] for item in batch])
    captions = [item[1] for item in batch]
    CLIP_tokens = torch.stack([item[2] for item in batch])
    keys = [item[3] for item in batch]
    return imgs, captions, CLIP_tokens, keys

train_dl = data.DataLoader(train_dataset, batch_size=2, collate_fn=collate_fn)
test_dl = data.DataLoader(test_dataset, batch_size=2, collate_fn=collate_fn)

# Create text encoder
text_encoder = SimpleTextEncoder(sent_dim=5, word_dim=6)

# Create args object
class Args:
    def __init__(self):
        self.device = 'cpu'
        self.z_dim = 7

args = Args()

# Define ARGS dict
ARGS = {
    'train_dl': train_dl,
    'test_dl': test_dl,
    'text_encoder': text_encoder,
    'args': args
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
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
get_one_batch_data Ran
ran on a synthesized input
oxygenlu/ratlip/code/lib/datasets.py
code served (permissive licence) · get_code("7092063b4f5817fa")
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 tuple. The image it ran in was not recorded. If your copy returns the same type 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
import torch.utils.data as data

torch.manual_seed(0)

# Stand-in text encoder that mimics the expected behavior
class MockTextEncoder(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(8, 16)
    
    def forward(self, caption):
        # caption shape: (batch_size, seq_len, embedding_dim) or similar
        # Returns: (sent_emb, words_embs)
        batch_size = caption.shape[0]
        sent_emb = torch.randn(batch_size, 16)
        words_embs = torch.randn(batch_size, 4, 16)
        return sent_emb, words_embs

# Stand-in args object
class Args:
    def __init__(self, device):
        self.device = device

# Create mock data batch
# Expected structure: (imgs, captions, CLIP_tokens, keys)
batch_imgs = torch.randn(2, 3, 4, 4)  # batch_size=2, channels=3, height=4, width=4
batch_captions = ["a dog", "a cat"]
batch_CLIP_tokens = torch.randint(0, 100, (2, 8))  # batch_size=2, seq_len=8
batch_keys = ["key1", "key2"]

mock_batch = (batch_imgs, batch_captions, batch_CLIP_tokens, batch_keys)

# Create dataloader wrapping the batch
dataloader = [mock_batch]

# Instantiate text encoder
text_encoder = MockTextEncoder()

# Create args with CPU device
args = Args(device='cpu')

ARGS = {
    'dataloader': dataloader,
    'text_encoder': text_encoder,
    'args': args
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
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
load_npz Ran
ran on a synthesized input
oxygenlu/ratlip/code/lib/utils.py
code served (permissive licence) · get_code("6364051ef9e418c9")
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 tuple. The image it ran in was not recorded. If your copy returns the same type 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 numpy as np
import tempfile
import os

# Create a temporary npz file with the required structure
temp_dir = tempfile.mkdtemp()
temp_path = os.path.join(temp_dir, "test_data.npz")

# Create sample data with distinct dimensions
mu_data = np.random.randn(4, 3).astype(np.float32)
sigma_data = np.random.randn(4, 5).astype(np.float32)

# Save as npz file
np.savez(temp_path, mu=mu_data, sigma=sigma_data)

# Define ARGS
ARGS = {
    "path": temp_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
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
maybe_min Ran
ran, honoured the contract we drafted
drboog/Lafite/dataset_tool.py
pointer only (licence: MIT); the stored copy get_code serves, from pdillis/stylegan3-fun, is pointer only (licence: NOASSERTION) · get_code("dd1dc700e87f36cc")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from pdillis/stylegan3-fun, and GitHub could not classify the licence of that repository (NOASSERTION). The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

params_count Ran
ran on a synthesized input
oxygenlu/ratlip/code/lib/utils.py
code served (permissive licence) · get_code("edd67ef305db54ef")
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 int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was 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
import numpy as np

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 8)
        self.linear2 = nn.Linear(8, 3)
    
    def forward(self, x):
        return self.linear2(self.linear1(x))

model = SimpleModel()

ARGS = {
    "model": model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () 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
prepare_data Ran
ran on a synthesized input
oxygenlu/ratlip/code/lib/datasets.py
code served (permissive licence) · get_code("31efd08e9e090d29")
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 tuple. The image it ran in was not recorded. If your copy returns the same type 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

torch.manual_seed(0)

# Create a minimal text encoder that matches the interface expected
class MinimalTextEncoder(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(10, 4)
    
    def forward(self, tokens):
        # tokens shape: (batch_size, seq_len, token_dim)
        # Return: (sent_emb, words_embs)
        # sent_emb: (batch_size, emb_dim)
        # words_embs: (batch_size, seq_len, emb_dim)
        batch_size = tokens.shape[0]
        sent_emb = self.linear(tokens.mean(dim=1))  # (batch_size, 4)
        words_embs = self.linear(tokens)  # (batch_size, seq_len, 4)
        return sent_emb, words_embs

# Create minimal test data
batch_size = 2
seq_len = 5
token_dim = 10
img_dim = 6

# Batch data: (imgs, captions, CLIP_tokens, keys)
imgs = torch.randn(batch_size, 3, img_dim)
captions = ["a cat sitting", "a dog running"]
CLIP_tokens = torch.randn(batch_size, seq_len, token_dim)
keys = ["img_1", "img_2"]

data = (imgs, captions, CLIP_tokens, keys)

text_encoder = MinimalTextEncoder()
device = "cpu"

ARGS = {
    "data": data,
    "text_encoder": text_encoder,
    "device": device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
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
MA_GP_MP Not yet run
unverified (harvested, no recorded run)
oxygenlu/ratlip/code/lib/modules.py
code served (permissive licence) · get_code("e881f3aa13c1dfca")
How it ran

no invocation recorded.

choose_model Not yet run
unverified (harvested, no recorded run)
oxygenlu/ratlip/code/lib/utils.py
code served (permissive licence) · get_code("1307498a08bac630")
How it ran

no invocation recorded.

get_imgs Not yet run
unverified (harvested, no recorded run)
drboog/Lafite/datasets/datasets_bird_coco.py
code served (permissive licence) · get_code("964f3c53ef8df7cd")
How it ran

no invocation recorded.

load_filenames Not yet run
unverified (harvested, no recorded run)
drboog/Lafite/datasets/datasets_bird_coco.py
code served (permissive licence) · get_code("2c903601913a4f24")
How it ran

no invocation recorded.

prepare_data Not yet run
unverified (harvested, no recorded run)
drboog/Lafite/datasets/datasets_bird_coco.py
code served (permissive licence) · get_code("5123f0670f4adce1")
How it ran

no invocation recorded.

prepare_dataset Not yet run
unverified (harvested, no recorded run)
oxygenlu/ratlip/code/lib/perpare.py
code served (permissive licence) · get_code("d4c40e264c45657c")
How it ran

no invocation recorded.

Also present in other papers' code

Of 18 samples lifted from this paper's linked repositories, 18 carry a function identity and 0 do not. Of the 18 distinct functions with an identity, 4 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 2 the same month, 4 later) and 14 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.

FunctionAlso present inEarliest-datedOthers named
file_ext 35 other papers
1 dated earlier, 1 the same month, 33 later
2106.07023 (2021-06)
Styleformer: Transformer based Generative Adversarial Networks with Style Vector
2111.02175 (2021-11)
Discriminator Synthesis: On reusing the other half of Generative Adversarial Networks
2201.06578 (2022-01)
Collapse by Conditioning: Training Class-conditional GANs with Limited Data
and 32 more not named here
maybe_min 35 other papers
1 dated earlier, 1 the same month, 33 later
2106.07023 (2021-06)
Styleformer: Transformer based Generative Adversarial Networks with Style Vector
2111.02175 (2021-11)
Discriminator Synthesis: On reusing the other half of Generative Adversarial Networks
2201.06578 (2022-01)
Collapse by Conditioning: Training Class-conditional GANs with Limited Data
and 32 more not named here
autopad 23 other papers
6 dated earlier, 17 later · 1 only as an assembly brick
1311.2524 (2013-11)
Rich feature hierarchies for accurate object detection and semantic segmentation
1703.07402 (2017-03)
Simple Online and Realtime Tracking with a Deep Association Metric
1804.02767 (2018-04)
YOLOv3: An Incremental Improvement
and 20 more not named here
get_G_in_out_chs 1 other paper
1 later
2301.12959 (2023-01)
GALIP: Generative Adversarial CLIPs for Text-to-Image Synthesis

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

One of the major challenges in training text-to-image generation models is the need of a large number of high-quality image-text pairs. While image samples are often easily accessible, the associated text descriptions typically require careful human captioning, which is particularly time- and cost-consuming. In this paper, we propose the first work to train text-to-image generation models without any text data. Our method leverages the well-aligned multi-modal semantic space of the powerful pre-trained CLIP model: the requirement of text-conditioning is seamlessly alleviated via generating text features from image features. Extensive experiments are conducted to illustrate the effectiveness of the proposed method. We obtain state-of-the-art results in the standard text-to-image generation tasks. Importantly, the proposed language-free model outperforms most existing models trained with full image-text pairs. Furthermore, our method can be applied in fine-tuning pre-trained models, which saves both training time and cost in training text-to-image generation models. Our pre-trained model obtains competitive results in zero-shot text-to-image generation on the MS-COCO dataset, yet with around only 1% of the model size and training data size relative to the recently proposed large DALL-E model.

For agents

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

get_harvested_code_for_paper("2111.13792")
get_code_for_paper("2111.13792")
have("2111.13792")

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

curl https://syntology.ai/api/ran/2111.13792.json

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

[![Syntology run record](https://syntology.ai/api/ran/2111.13792.svg)](https://syntology.ai/paper/2111.13792)

Connect an agent — have() is free.