SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2605.26283 · 2026

Benchmarking Convolutional, Transformer, Hybrid, and Vision Language Models for Multi Disease Retinal Screening

Durjoy Dey

arXiv · PDF · Open in the Atlas

Code that ran

10 of 17 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; 9 failures of our instrument; 7 no recorded run), measured 2026-11-22; 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/2605.26283.json · badge: /api/ran/2605.26283.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
Durjoy001/Retinal-NeuralNET canonical (classifier label)
classifier basis: census
10 ran · 7 no recorded run
FunctionStatusWhere it lives
build_splits_by_nodule Ran
ran; our contract draft was wrong, not the code
Durjoy001/Retinal-NeuralNET/misc/train_vit.py
code served (permissive licence) · get_code("24e0e309b129e625")
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 h5py
import numpy as np
import pandas as pd
import tempfile
import os

# Create a temporary HDF5 file with the required structure
temp_dir = tempfile.mkdtemp()
h5_path = os.path.join(temp_dir, "test_data.h5")

with h5py.File(h5_path, 'w') as h5f:
    # Create annot_id dataset with some test IDs
    annot_ids = np.array([b'nodule_001', b'nodule_002', b'nodule_003', b'nodule_004', b'nodule_005'], dtype='S20')
    h5f.create_dataset('annot_id', data=annot_ids)

# Create a metadata DataFrame with matching annot_ids and ti-rads_composition labels
meta_df = pd.DataFrame({
    'annot_id': ['nodule_001', 'nodule_002', 'nodule_003', 'nodule_004', 'nodule_005'],
    'ti-rads_composition': [1, 2, 1, 3, 2]
})

ARGS = {
    'h5_path': h5_path,
    'meta_df': meta_df
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
h5py, numpy, pandas — 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
compute_confusion_from_preds Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Durjoy001/Retinal-NeuralNET/misc/train_cnn_final.py
code served (permissive licence) · get_code("c55602a82bda33e7")
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, ndim 1, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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

# Create small binary classification test data
np.random.seed(0)

# y_true: ground truth binary labels (0 or 1)
y_true = np.array([0, 1, 1, 0, 1, 1, 0, 0])

# y_pred_bin: predicted binary labels (0 or 1)
y_pred_bin = np.array([0, 1, 0, 0, 1, 1, 0, 1])

ARGS = {
    "y_true": y_true,
    "y_pred_bin": y_pred_bin,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family int — a tuple output has no array shape of its own; none is stated
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
compute_sens_spec Ran
ran; our contract draft was wrong, not the code
Durjoy001/Retinal-NeuralNET/src/train_cnn.py
code served (permissive licence) · get_code("3e2c94e6bdd139f3")
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, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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)

# Create small binary prediction and label tensors
# Shape: (batch_size=2, num_classes=4)
# preds should be binary (0 or 1)
# labels should be binary (0 or 1)
preds = torch.tensor([
    [1, 0, 1, 0],
    [0, 1, 1, 0]
], dtype=torch.float32)

labels = torch.tensor([
    [1, 0, 0, 1],
    [0, 1, 1, 0]
], dtype=torch.float32)

ARGS = {
    'preds': preds,
    'labels': labels,
    'average': 'macro'
}
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, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
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
cosine_with_warmup Ran
ran; our contract draft was wrong, not the code
Durjoy001/Retinal-NeuralNET/src/train_hybrid.py
code served (permissive licence) · get_code("950fd803ff99e7b3")
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 LambdaLR. 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 random, warnings, numpy as np, pandas as pd, math
import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a simple model with parameters
model = nn.Linear(4, 3)

# Create an optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=0.1)

# Define ARGS with all parameters for cosine_with_warmup
ARGS = {
    "optimizer": optimizer,
    "warmup_epochs": 5,
    "total_epochs": 20
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LambdaLR — a LambdaLR output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LambdaLR 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
cosine_with_warmup Ran
ran; our contract draft was wrong, not the code
Durjoy001/Retinal-NeuralNET/src/train_vlm.py
code served (permissive licence) · get_code("24be0e1ac502e565")
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 LambdaLR. 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 math

torch.manual_seed(0)

# Create a simple model
model = nn.Linear(4, 3)

# Create an optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=0.1)

# Define ARGS dictionary with all parameters
ARGS = {
    "optimizer": optimizer,
    "warmup_epochs": 5,
    "total_epochs": 20,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LambdaLR — a LambdaLR output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a LambdaLR 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
evaluate_with_preds Ran
ran; our contract draft was wrong, not the code
Durjoy001/Retinal-NeuralNET/misc/train_vit.py
code served (permissive licence) · get_code("e854d8591c60b644")
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 numpy as np

torch.manual_seed(0)

# Define a minimal model that satisfies the function's usage
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)  # 4 input features, 3 classes
    
    def forward(self, x):
        return self.linear(x)

# Create model instance
model = SimpleModel()

# Create a simple data loader as a list of (imgs, labels) tuples
# Each img batch is shape (2, 4), labels are shape (2,)
loader = [
    (torch.randn(2, 4), torch.tensor([0, 1])),
    (torch.randn(2, 4), torch.tensor([2, 0])),
]

# Device - CPU only
device = 'cpu'

ARGS = {
    'model': model,
    'loader': loader,
    'device': device,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, sklearn, torch, tqdm — 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_transforms Ran
ran; our contract draft was wrong, not the code
Durjoy001/Retinal-NeuralNET/src/train_cnn.py
code served (permissive licence) · get_code("c107b9b21054d659")
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 Compose. 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
from torchvision import transforms, models
from torchvision.models import DenseNet121_Weights, ResNet50_Weights, EfficientNet_B3_Weights, Inception_V3_Weights

torch.manual_seed(0)

ARGS = {
    'model_name': 'resnet50',
    'train': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Compose — a Compose output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Compose 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_transforms Ran
ran; our contract draft was wrong, not the code
Durjoy001/Retinal-NeuralNET/misc/train_cnn_final.py
code served (permissive licence) · get_code("839ba542f6d0ce23")
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 Compose. 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 numpy as np
from PIL import Image
from torchvision import transforms

torch.manual_seed(0)

# Create a small test image (PIL Image)
# Make it large enough for CenterCrop(1016) to work
test_image_array = np.random.randint(0, 256, (1200, 1200, 3), dtype=np.uint8)
test_image = Image.fromarray(test_image_array)

ARGS = {
    "model_name": "resnet50",
    "train": True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Compose — a Compose output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Compose 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
train_one_epoch Ran
ran, honoured the contract we drafted
Durjoy001/Retinal-NeuralNET/misc/train_vit.py
code served (permissive licence) · get_code("adfe9e91da21b6e1")
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, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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)

# Build a simple model
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

model = SimpleModel()

# Build a small data loader: list of (imgs, labels) tuples
# batch_size=2, img_dim=4, num_classes=3
loader = [
    (torch.randn(2, 4), torch.randint(0, 3, (2,))),
    (torch.randn(2, 4), torch.randint(0, 3, (2,))),
]

# Loss function for multiclass classification
criterion = nn.CrossEntropyLoss()

# Optimizer
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

# Device
device = 'cpu'

ARGS = {
    'model': model,
    'loader': loader,
    'criterion': criterion,
    'optimizer': optimizer,
    'device': device,
}
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 tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
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
vlm_transforms Ran
ran; our contract draft was wrong, not the code
Durjoy001/Retinal-NeuralNET/src/train_vlm.py
code served (permissive licence) · get_code("1e20b073f2b029d9")
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 Compose. 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 numpy as np
from torchvision import transforms
from torchvision.transforms import InterpolationMode
import timm.data

torch.manual_seed(0)

# Test with a CLIP model name
ARGS = {
    "model_name": "clip_vit_base_patch32_224",
    "train": True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, timm, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Compose — a Compose output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Compose 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
build_model Not yet run
unverified (harvested, no recorded run)
Durjoy001/Retinal-NeuralNET/src/train_cnn.py
code served (permissive licence) · get_code("9a15b6f14e5e71e6")
How it ran

no invocation recorded.

build_model Not yet run
unverified (harvested, no recorded run)
Durjoy001/Retinal-NeuralNET/src/train_hybrid.py
code served (permissive licence) · get_code("a4dfe1619a248e9e")
How it ran

no invocation recorded.

build_model Not yet run
unverified (harvested, no recorded run)
Durjoy001/Retinal-NeuralNET/src/train_vit.py
code served (permissive licence) · get_code("8ba2159b6ca06dd7")
How it ran

no invocation recorded.

build_model Not yet run
unverified (harvested, no recorded run)
Durjoy001/Retinal-NeuralNET/misc/train_cnn_final.py
code served (permissive licence) · get_code("4e4ae5a8d5f164fd")
How it ran

no invocation recorded.

build_model Not yet run
unverified (harvested, no recorded run)
Durjoy001/Retinal-NeuralNET/misc/train_vit_update.py
code served (permissive licence) · get_code("c393ea4a983d138f")
How it ran

no invocation recorded.

vit_transforms Not yet run
unverified (harvested, no recorded run)
Durjoy001/Retinal-NeuralNET/src/train_hybrid.py
code served (permissive licence) · get_code("1d022956c4f0102c")
How it ran

no invocation recorded.

vit_transforms Not yet run
unverified (harvested, no recorded run)
Durjoy001/Retinal-NeuralNET/src/train_vit.py
code served (permissive licence) · get_code("78641b41e464f9ab")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 16 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 16 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

Modern deep learning offers powerful tools for automated retinal screening, but it remains unclear how different visual model families compare in realistic multi-disease settings and under domain shift. In this work, we benchmark twelve architectures across four model families: convolutional neural networks (CNNs), vision transformers (ViTs), hybrid CNN-transformer backbones, and vision-language models (VLMs), using the Retinal Fundus Multidisease Image Dataset (RFMiD), a multi-label fundus dataset with 28 disease classes. We evaluate two tasks relevant to screening practice: (i) a binary screening task that determines whether an image contains any retinal disease, and (ii) a multi-label classification task, which predicts the full spectrum of RFMiD pathologies present in each image. Using standardized family-specific training configurations and a unified calibration and evaluation protocol, we report AUC, F1, precision, recall, and sensitivity at a clinically relevant operating point with specificity near 80% across all models. On RFMiD, all architectures perform well on the binary screening task (AUC above 84%), but attention-based models dominate: SwinTiny and the hybrid CoAtNet0 and MaxViTTiny achieve the highest AUC and F1, and significantly improve macro and micro F1 in the more challenging multi-label setting. Vision-language models (CLIP ViT-B/16 and SigLIP-Base384) are competitive with CNN baselines but do not surpass the best transformer and hybrid backbones. In external validation on Messidor-2 for referable diabetic retinopathy (DR), AUC ranges from 66.8% to 84.7%, with hybrid and transformer models again showing the strongest performance. The SigLIP-Base384 vision-language model demonstrates favorable precision-recall trade-offs for referable DR. To our knowledge, this is the first study to benchmark convolutional, transformer, hybrid, and vision-language models on RFMiD using a unified evaluation protocol with external validation on Messidor-2. Our results provide a reproducible reference for model choice in multi-disease retinal screening and guide the development of future automated screening tools for clinical deployment. • Computing methodologies → Machine learning; Computer vision; • Applied computing → Health care information systems.

For agents

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

get_harvested_code_for_paper("2605.26283")
get_code_for_paper("2605.26283")
have("2605.26283")

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

curl https://syntology.ai/api/ran/2605.26283.json

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

[![Syntology run record](https://syntology.ai/api/ran/2605.26283.svg)](https://syntology.ai/paper/2605.26283)

Connect an agent — have() is free.