SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1806.07421 · 2018

RISE: Randomized Input Sampling for Explanation of Black-box Models

arXiv · PDF · Open in the Atlas

Code that ran

32 of 34 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 28 have no contract verdict; 2 failures of our instrument; 2 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/1806.07421.json · badge: /api/ran/1806.07421.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
ftorres11/saliencysense community (classifier label)
classifier basis not recorded on the link
16 ran · 1 no recorded run
openvinotoolkit/openvino_xai community (classifier label)
classifier basis not recorded on the link
8 ran · 0 no recorded run
vlue-c/PyTorch-Explanations community (classifier label)
classifier basis not recorded on the link
6 ran · 1 no recorded run
openvinotoolkit/datumaro community (classifier label)
classifier basis not recorded on the link
2 ran · 0 no recorded run
FunctionStatusWhere it lives
binary_gumbel_softmax Ran
ran on a synthesized input
ftorres11/saliencysense/casme/archs.py
code served (permissive licence) · get_code("c75a899b50d166b6")
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 (2, 4) 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

torch.manual_seed(0)

# Construct arguments for binary_gumbel_softmax
logits = torch.randn(2, 4)  # Small batch of logits
tau = 0.1
hard = False
epsilon = 1e-08

ARGS = {
    "logits": logits,
    "tau": tau,
    "hard": hard,
    "epsilon": epsilon
}
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 2, dtype family float, shape (2, 4) — shape (2, 4) 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
cam_targetter Ran
ran on a synthesized input
ftorres11/saliencysense/models/utils.py
code served (permissive licence) · get_code("3ff2de20bf374daa")
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 list, 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 minimal ResNet-like model
class MinimalResNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.layer4 = nn.Sequential(
            nn.Conv2d(3, 64, 3),
            nn.Conv2d(64, 64, 3)
        )
    
    def forward(self, x):
        return self.layer4(x)

# Build a minimal VGG-like model
class MinimalVGG(nn.Module):
    def __init__(self):
        super().__init__()
        self.features = nn.Sequential(
            nn.Conv2d(3, 64, 3),
            nn.Conv2d(64, 64, 3)
        )
    
    def forward(self, x):
        return self.features(x)

# Build a minimal ConvNeXt-like model
class MinimalConvNeXt(nn.Module):
    def __init__(self):
        super().__init__()
        self.features = nn.Sequential(
            nn.Sequential(
                nn.Conv2d(3, 64, 3),
                nn.Sequential(
                    nn.Conv2d(64, 64, 3),
                    nn.Conv2d(64, 64, 3),
                    nn.Conv2d(64, 64, 3),
                    nn.Conv2d(64, 64, 3),
                    nn.Conv2d(64, 64, 3),
                    nn.Conv2d(64, 64, 3)
                )
            )
        )
    
    def forward(self, x):
        return self.features(x)

# Build a minimal ViT-like model
class MinimalVisionTransformer(nn.Module):
    def __init__(self):
        super().__init__()
        self.blocks = nn.ModuleList([
            nn.ModuleDict({
                'norm1': nn.LayerNorm(64)
            }),
            nn.ModuleDict({
                'norm1': nn.LayerNorm(64)
            })
        ])
    
    def forward(self, x):
        return x

# Test with ResNet
model_resnet = MinimalResNet()
ARGS = {"model": model_resnet}
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 list, ndim 1, dtype family float — a list 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
colormap Ran
ran on a synthesized input
openvinotoolkit/openvino_xai/openvino_xai/explainer/visualizer.py
pointer only (licence: Apache-2.0) · get_code("ad36990c6e8e7653")
How it ran

The body of this sample is withheld: Apache-2.0, as GitHub detects it on this paper's repository, openvinotoolkit/openvino_xai, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

convert_targets_to_numpy Ran
ran on a synthesized input
openvinotoolkit/openvino_xai/openvino_xai/explainer/utils.py
pointer only (licence: Apache-2.0) · get_code("e3c80d43208cb814")
How it ran

The body of this sample is withheld: Apache-2.0, as GitHub detects it on this paper's repository, openvinotoolkit/openvino_xai, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

default_transform Ran
ran on a synthesized input
vlue-c/PyTorch-Explanations/torchvex/cam/hook.py
code served (permissive licence) · get_code("3d00a41f874efe78")
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 (2, 4) 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

torch.manual_seed(0)

# Test case 1: torch.Tensor input
tensor_input = torch.randn(2, 4)

# Test case 2: non-tensor input (list, tuple, or other indexable object)
list_input = [torch.randn(2, 4), torch.randn(2, 4)]

ARGS = {
    "feature": tensor_input
}
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 2, dtype family float, shape (2, 4) — shape (2, 4) 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
explains_all Ran
ran on a synthesized input
openvinotoolkit/openvino_xai/openvino_xai/explainer/utils.py
pointer only (licence: Apache-2.0) · get_code("b88181348223ac7a")
How it ran

The body of this sample is withheld: Apache-2.0, as GitHub detects it on this paper's repository, openvinotoolkit/openvino_xai, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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_gaussian_filter Ran
ran on a synthesized input
vlue-c/PyTorch-Explanations/torchvex/metrics/causal.py
code served (permissive licence) · get_code("98736d68924d20ac")
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 (5, 5) 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
from numbers import Number
from scipy.ndimage.filters import gaussian_filter

torch.manual_seed(0)

ARGS = {
    "size": 5,
    "sig": 1.0
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
scipy, torch — 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 (5, 5) — shape (5, 5) 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
get_mask Ran
ran on a synthesized input
ftorres11/saliencysense/casme/model_basics.py
code served (permissive licence) · get_code("a2f9db8c0650f2df")
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 (2, 4) 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

torch.manual_seed(0)

# Create a minimal classifier that returns intermediate layers
class MinimalClassifier(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc1 = nn.Linear(4, 8)
        self.fc2 = nn.Linear(8, 5)
    
    def forward(self, x, return_intermediate=False):
        layer1 = self.fc1(x)
        output = self.fc2(layer1)
        if return_intermediate:
            return output, [layer1]
        return output

# Create a minimal masker that accepts layers and returns masks
class MinimalMasker(nn.Module):
    def __init__(self):
        super().__init__()
        self.mask_layer = nn.Linear(8, 4)
    
    def forward(self, layers, use_p=None, class_ids=None, no_sigmoid=False):
        # layers is a list of intermediate tensors
        x = layers[0]  # Use first intermediate layer
        mask = self.mask_layer(x)
        if not no_sigmoid:
            mask = torch.sigmoid(mask)
        return mask

# Build the model dict with both classifier and masker
classifier = MinimalClassifier()
masker = MinimalMasker()
model = {
    'classifier': classifier,
    'masker': masker
}

# Create input tensor (batch size 2, feature dim 4)
input_tensor = torch.randn(2, 4)

# Define all arguments to the function
ARGS = {
    'input_': input_tensor,
    'model': model,
    'use_p': None,
    'class_ids': None,
    'get_output': False,
    'no_sigmoid': False
}
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 2, dtype family float, shape (2, 4) — shape (2, 4) 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
get_masked_images Ran
ran on a synthesized input
ftorres11/saliencysense/casme/casme_utils.py
code served (permissive licence) · get_code("544506b6b4e94b6e")
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 5, 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 test tensors
# images: shape (batch=2, channels=3, height=4, width=4)
images = torch.randn(2, 3, 4, 4)

# binary_mask: same shape as images for element-wise multiplication
binary_mask = torch.randint(0, 2, (2, 3, 4, 4)).float()

# gray_scale: integer parameter controlling behavior
gray_scale = 1

# return_mask: boolean parameter controlling return value
return_mask = True

ARGS = {
    "images": images,
    "binary_mask": binary_mask,
    "gray_scale": gray_scale,
    "return_mask": return_mask,
}
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 5, 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
get_masked_images_v2 Ran
ran on a synthesized input
ftorres11/saliencysense/casme/casme_utils.py
code served (permissive licence) · get_code("b257e814f0b02a0e")
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 3, 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 batch tensors
batch_x = torch.randn(2, 4)  # batch_size=2, features=4
mask = torch.randint(0, 2, (2, 4)).float()  # binary mask with same shape

ARGS = {
    "batch_x": batch_x,
    "mask": mask,
}
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 3, 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
get_masks_and_check_predictions Ran
ran on a synthesized input
ftorres11/saliencysense/casme/model_basics.py
code served (permissive licence) · get_code("1eef12cb27f7130e")
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. 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 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 SimpleClassifier(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(48, 10)
        self.layer1 = nn.Linear(48, 32)
        self.layer2 = nn.Linear(32, 16)
    
    def forward(self, x, return_intermediate=False):
        batch_size = x.size(0)
        x_flat = x.view(batch_size, -1)
        layer1_out = self.layer1(x_flat)
        layer2_out = self.layer2(layer1_out)
        output = self.fc(x_flat)
        
        if return_intermediate:
            layers = [layer1_out, layer2_out]
            return output, layers
        return output

class SimpleMasker(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(48, 224 * 224)
    
    def forward(self, layers, use_p=None, class_ids=None, no_sigmoid=False):
        batch_size = layers[0].size(0)
        x = torch.cat([layers[0], layers[1]], dim=1)
        mask_flat = self.fc(x)
        mask = mask_flat.view(batch_size, 1, 224, 224)
        
        if not no_sigmoid:
            mask = torch.sigmoid(mask)
        
        return mask

classifier = SimpleClassifier()
masker = SimpleMasker()

model = {
    'classifier': classifier,
    'masker': masker
}

input_tensor = torch.randn(2, 1, 4, 12)
target_tensor = torch.tensor([3, 7], dtype=torch.long)

ARGS = {
    'input_': input_tensor,
    'target': target_tensor,
    'model': model,
    'erode_k': 0,
    'dilate_k': 0,
    'use_p': None,
    'no_sigmoid': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, 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_min_max Ran
ran on a synthesized input
openvinotoolkit/openvino_xai/openvino_xai/common/utils.py
pointer only (licence: Apache-2.0) · get_code("20fcc158d33e1d1c")
How it ran

The body of this sample is withheld: Apache-2.0, as GitHub detects it on this paper's repository, openvinotoolkit/openvino_xai, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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_target_indices Ran
ran on a synthesized input
openvinotoolkit/openvino_xai/openvino_xai/explainer/utils.py
pointer only (licence: Apache-2.0) · get_code("911b3dde6c0f7336")
How it ran

The body of this sample is withheld: Apache-2.0, as GitHub detects it on this paper's repository, openvinotoolkit/openvino_xai, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

ifnone Ran
ran, honoured the contract we drafted
ftorres11/saliencysense/lib/IBA_utils.py
code served (permissive licence) · get_code("6f4763ec77290638")
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 numpy as np

torch.manual_seed(0)

ARGS = {
    "a": None,
    "b": 42
}
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
ignore_superclass_signature Ran
ran on a synthesized input
ftorres11/saliencysense/casme/core.py
code served (permissive licence) · get_code("9c25a48e312acd33")
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 function. 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

def ignore_superclass_signature(f):
    return f

# The function simply returns its argument unchanged
# It takes a single parameter 'f' which should be a callable (function)
# We'll create a simple function to pass as the argument

def dummy_function(x):
    """A simple dummy function to pass as argument"""
    return x + 1

ARGS = {
    'f': dummy_function
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
imagenet_tester Ran
ran on a synthesized input
ftorres11/saliencysense/lib/data.py
code served (permissive licence) · get_code("5591ec5f001d8f12")
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

torch.manual_seed(0)

# The function imagenet_tester takes two parameters:
# - resize: an integer for resizing the image
# - crop_size: an integer for center cropping the image
# Both are used as integers in the transforms pipeline

ARGS = {
    "resize": 256,
    "crop_size": 224,
}
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
imagenet_transform Ran
ran on a synthesized input
vlue-c/PyTorch-Explanations/torchvex/utils/dataset.py
code served (permissive licence) · get_code("aa308fe2fc8d575e")
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
from torchvision import transforms as T
from PIL import Image
import numpy as np

torch.manual_seed(0)

# Create a small dummy image for testing the transform
dummy_image_array = np.random.randint(0, 256, (224, 224, 3), dtype=np.uint8)
dummy_image = Image.fromarray(dummy_image_array)

ARGS = {
    "centercrop": 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
insert_into_sequential Ran
ran; our contract draft was wrong, not the code
ftorres11/saliencysense/lib/IBA.py
code served (permissive licence) · get_code("2051b0d0f5da07d3")
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 Sequential. 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 simple sequential model with a few layers
sequential = nn.Sequential(
    nn.Linear(4, 3),
    nn.ReLU(),
    nn.Linear(3, 2)
)

# Create a layer to insert
layer = nn.Dropout(0.5)

# Position to insert at
idx = 1

ARGS = {
    "sequential": sequential,
    "layer": layer,
    "idx": idx
}
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 Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential output the recorded ndim 1 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
loglevel Ran
ran on a synthesized input
openvinotoolkit/datumaro/src/datumaro/cli/__main__.py
pointer only (licence: MIT) · get_code("28e1a63b82c1d344")
How it ran

The body of this sample is withheld: MIT, as GitHub detects it on this paper's repository, openvinotoolkit/datumaro, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

make_blurred_circular_mask Ran
ran on a synthesized input
vlue-c/PyTorch-Explanations/torchvex/meaningful_perturbation/mask.py
code served (permissive licence) · get_code("a317e3f45ccf837b")
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 (35, 1, 4, 3) 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

torch.manual_seed(0)

# Create a small image tensor
# The function expects image to have shape (..., H, W)
# We'll create a batch of 2 images with height 4 and width 3
image = torch.randn(2, 3, 4, 3)

# Create radius values
rads = torch.arange(0, 175, 5)

ARGS = {
    "image": image,
    "rads": rads
}
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 (35, 1, 4, 3) — shape (35, 1, 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
make_subset_from_fnames Ran
ran on a synthesized input
vlue-c/PyTorch-Explanations/torchvex/utils/dataset.py
code served (permissive licence) · get_code("51ae8cf08b64ff6f")
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 Subset. 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 os
from torch.utils.data import Subset, Dataset

class StandInDataset(Dataset):
    def __init__(self):
        # samples is a list of tuples (path, label)
        self.samples = [
            ("/path/to/image_001.jpg", 0),
            ("/path/to/image_002.jpg", 1),
            ("/another/dir/image_003.jpg", 0),
            ("/another/dir/image_004.jpg", 1),
        ]
    
    def __len__(self):
        return len(self.samples)
    
    def __getitem__(self, idx):
        return self.samples[idx]

dset = StandInDataset()
fnames = ["image_001.jpg", "image_003.jpg"]

ARGS = {
    "dset": dset,
    "fnames": fnames,
}
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 Subset — a Subset output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Subset output the recorded ndim 2 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
outlier_deprocessor Ran
ran on a synthesized input
ftorres11/saliencysense/lib/data.py
code served (permissive licence) · get_code("c65481685dc3709c")
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, 4, 3, 5) 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

torch.manual_seed(0)

# The function takes grad_batch as input
# It performs operations: mean, std on dims (2,3), normalization, scaling, and clamping
# grad_batch should be a 4D tensor (batch, channels, height, width)
grad_batch = torch.randn(2, 4, 3, 5)

ARGS = {
    "grad_batch": grad_batch
}
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, 4, 3, 5) — shape (2, 4, 3, 5) 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
overlay Ran
ran on a synthesized input
openvinotoolkit/openvino_xai/openvino_xai/explainer/visualizer.py
pointer only (licence: Apache-2.0) · get_code("4d8d016663759432")
How it ran

The body of this sample is withheld: Apache-2.0, as GitHub detects it on this paper's repository, openvinotoolkit/openvino_xai, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

remove_plugin_type Ran
ran on a synthesized input
openvinotoolkit/datumaro/src/datumaro/components/cli_plugin.py
pointer only (licence: MIT) · get_code("09c5cbd7a761b5cc")
How it ran

The body of this sample is withheld: MIT, as GitHub detects it on this paper's repository, openvinotoolkit/datumaro, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

replace_prefix Ran
ran on a synthesized input
ftorres11/saliencysense/casme/archs.py
code served (permissive licence) · get_code("afc608bf21aa9f05")
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 str. 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

ARGS = {
    "s": "prefix_test_string",
    "prefix": "prefix_"
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
resize Ran
ran on a synthesized input
openvinotoolkit/openvino_xai/openvino_xai/explainer/visualizer.py
pointer only (licence: Apache-2.0) · get_code("eabd217e202d8b8b")
How it ran

The body of this sample is withheld: Apache-2.0, as GitHub detects it on this paper's repository, openvinotoolkit/openvino_xai, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

resnet50shared Ran
ran on a synthesized input
ftorres11/saliencysense/casme/archs.py
code served (permissive licence) · get_code("f067739c23018f46")
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 ResNet. 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)

ARGS = {
    "pretrained": False,
    "path": None,
}
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 ResNet — a ResNet output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ResNet 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
scaling Ran
ran on a synthesized input
openvinotoolkit/openvino_xai/openvino_xai/common/utils.py
pointer only (licence: Apache-2.0) · get_code("161696665a9b902a")
How it ran

The body of this sample is withheld: Apache-2.0, as GitHub detects it on this paper's repository, openvinotoolkit/openvino_xai, is in the permissive set, but a licence decision keeps the body a pointer: the reason for that decision is not recorded with it. 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.

smooth_deprocessor Ran
ran on a synthesized input
ftorres11/saliencysense/lib/data.py
code served (permissive licence) · get_code("cb904fc4d37fb3ae")
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, 1, 3, 5) 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 sys

torch.manual_seed(0)

# Create a small batch of gradients
# Shape: (batch_size, channels, height, width)
# Using distinct dimensions: batch=2, channels=4, height=3, width=5
grad_batch = torch.randn(2, 4, 3, 5)

ARGS = {
    "grad_batch": grad_batch
}
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, 1, 3, 5) — shape (2, 1, 3, 5) 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
tensor_to_np_img Ran
ran; our contract draft was wrong, not the code
ftorres11/saliencysense/lib/IBA.py
code served (permissive licence) · get_code("62b5772d8033fdab")
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 ndarray, ndim 3, dtype family float, shape (4, 5, 3) 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 numpy as np

torch.manual_seed(0)

# The function takes img_t as input, which should be a torch tensor of shape (c, h, w)
# where c is channels, h is height, w is width
# The function expects a tensor with 3 channels (RGB), and applies normalization transforms

img_t = torch.randn(3, 4, 5)  # (channels=3, height=4, width=5)

ARGS = {
    "img_t": img_t
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family float, shape (4, 5, 3) — shape (4, 5, 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
to_unit_interval Ran
ran, honoured the contract we drafted
ftorres11/saliencysense/lib/IBA_utils.py
code served (permissive licence) · get_code("e5e556749746630d")
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 (2, 4) 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

torch.manual_seed(0)

# The function takes a single argument x and calls:
# - x.min() 
# - x.max()
# - arithmetic operations (subtraction, division)
# This means x should be a torch tensor with numeric values

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
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 2, dtype family float, shape (2, 4) — shape (2, 4) 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
total_variation Ran
ran on a synthesized input
vlue-c/PyTorch-Explanations/torchvex/meaningful_perturbation/mask.py
code served (permissive licence) · get_code("b0396293b38da996")
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. 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 total variation by taking differences along spatial dimensions
# It expects a tensor with shape (..., height, width)
# The function uses indexing like [..., 1:, :] and [..., :, 1:] for spatial differences
# So we need at least a 2D spatial tensor, but batch dimensions are allowed

# Create a small 4D tensor: (batch, channels, height, width)
inputs = torch.randn(2, 3, 4, 5)

# beta is a numeric parameter for the power operation
beta = 1

ARGS = {
    "inputs": inputs,
    "beta": beta,
}
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
gaussian_filter Not yet run
unverified (harvested, no recorded run)
vlue-c/PyTorch-Explanations/torchvex/meaningful_perturbation/mask.py
code served (permissive licence) · get_code("049ca000aec357e1")
How it ran

no invocation recorded.

load_monkeys Not yet run
unverified (harvested, no recorded run)
ftorres11/saliencysense/lib/IBA_utils.py
code served (permissive licence) · get_code("9a306d4046aceb10")
How it ran

no invocation recorded.

Also present in other papers' code

Of 34 samples lifted from this paper's linked repositories, 34 carry a function identity and 0 do not. Of the 34 distinct functions with an identity, 13 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 13 later) and 21 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
ifnone 2 other papers
2 later
2001.00396 (2020-01)
Restricting the Flow: Information Bottlenecks for Attribution
2104.00411 (2021-04)
Explaining COVID-19 and Thoracic Pathology Model Predictions by Identifying Informative Input Features
binary_gumbel_softmax 1 other paper
1 later
2010.09750 (2020-10)
Investigating and Simplifying Masking-based Saliency Methods for Model Interpretability
get_mask 1 other paper
1 later
2010.09750 (2020-10)
Investigating and Simplifying Masking-based Saliency Methods for Model Interpretability
get_masked_images 1 other paper
1 later
2010.09750 (2020-10)
Investigating and Simplifying Masking-based Saliency Methods for Model Interpretability
get_masked_images_v2 1 other paper
1 later
2010.09750 (2020-10)
Investigating and Simplifying Masking-based Saliency Methods for Model Interpretability
get_masks_and_check_predictions 1 other paper
1 later
2010.09750 (2020-10)
Investigating and Simplifying Masking-based Saliency Methods for Model Interpretability
ignore_superclass_signature 1 other paper
1 later
2010.09750 (2020-10)
Investigating and Simplifying Masking-based Saliency Methods for Model Interpretability
insert_into_sequential 1 other paper
1 later
2104.00411 (2021-04)
Explaining COVID-19 and Thoracic Pathology Model Predictions by Identifying Informative Input Features
load_monkeys 1 other paper
1 later
2104.00411 (2021-04)
Explaining COVID-19 and Thoracic Pathology Model Predictions by Identifying Informative Input Features
replace_prefix 1 other paper
1 later
2010.09750 (2020-10)
Investigating and Simplifying Masking-based Saliency Methods for Model Interpretability
resnet50shared 1 other paper
1 later
2010.09750 (2020-10)
Investigating and Simplifying Masking-based Saliency Methods for Model Interpretability
tensor_to_np_img 1 other paper
1 later
2104.00411 (2021-04)
Explaining COVID-19 and Thoracic Pathology Model Predictions by Identifying Informative Input Features
to_unit_interval 1 other paper
1 later
2104.00411 (2021-04)
Explaining COVID-19 and Thoracic Pathology Model Predictions by Identifying Informative Input Features

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

Deep neural networks are being used increasingly to automate data analysis and decision making, yet their decision-making process is largely unclear and is difficult to explain to the end users. In this paper, we address the problem of Explainable AI for deep neural networks that take images as input and output a class probability. We propose an approach called RISE that generates an importance map indicating how salient each pixel is for the model's prediction. In contrast to white-box approaches that estimate pixel importance using gradients or other internal network state, RISE works on black-box models. It estimates importance empirically by probing the model with randomly masked versions of the input image and obtaining the corresponding outputs. We compare our approach to state-of-the-art importance extraction methods using both an automatic deletion/insertion metric and a pointing metric based on human-annotated object segments. Extensive experiments on several benchmark datasets show that our approach matches or exceeds the performance of other methods, including white-box approaches. Project page: http://cs-people.bu.edu/vpetsiuk/rise/

For agents

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

get_harvested_code_for_paper("1806.07421")
get_code_for_paper("1806.07421")
have("1806.07421")

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

curl https://syntology.ai/api/ran/1806.07421.json

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

[![Syntology run record](https://syntology.ai/api/ran/1806.07421.svg)](https://syntology.ai/paper/1806.07421)

Connect an agent — have() is free.