SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.06197 · CVPR · 2024

Efficient Deformable ConvNets: Rethinking Dynamic and Sparse Operator for Vision Applications

Hongsheng Li, Yuwen Xiong, Xizhou Zhu, Lewei Lu, Jifeng Dai, Feng Wang, Yu Qiao, Wenhai Wang, Jie Zhou, Tong Lu, Zhiqi Li, Yuntao Chen, and 1 more

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 (2 of them constructed an object rather than computing a result; 4 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 4 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/2401.06197.json · badge: /api/ran/2401.06197.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
opengvlab/dcnv4 canonical (classifier label)
classifier basis: census
7 ran · 5 no recorded run
chenller/mmseg-extension community
Papers with Code archive
5 ran · 1 no recorded run
FunctionStatusWhere it lives
CenterFeatureScaleModule Ran
ran on a synthesized input
chenller/mmseg-extension/mmsegextlib/dcnv4/mmsegextlib_dcnv4/modules/dcnv4.py
code served (permissive licence) · get_code("d82968bff87f113b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, 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
import torch.nn as nn

# CenterFeatureScaleModule has no __init__ parameters (inherits from nn.Module only)
INIT_ARGS = {}

# The forward method takes query, center_feature_scale_proj_weight, and center_feature_scale_proj_bias
# query: input tensor, shape (batch, seq_len, d_model) or similar
# center_feature_scale_proj_weight: weight matrix for linear projection
# center_feature_scale_proj_bias: bias vector for linear projection

# Let's use:
# query shape: (2, 5, 8) - batch=2, seq_len=5, d_model=8
# weight shape: (6, 8) - projects from d_model=8 to output_dim=6
# bias shape: (6,) - output_dim=6

batch_size = 2
seq_len = 5
d_model = 8
output_dim = 6

query = torch.randn(batch_size, seq_len, d_model)
center_feature_scale_proj_weight = torch.randn(output_dim, d_model)
center_feature_scale_proj_bias = torch.randn(output_dim)

ARGS = {
    'query': query,
    'center_feature_scale_proj_weight': center_feature_scale_proj_weight,
    'center_feature_scale_proj_bias': center_feature_scale_proj_bias
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
DCNv4 Ran
ran on a synthesized input · constructed an object; no result was computed
chenller/mmseg-extension/mmsegextlib/dcnv4/mmsegextlib_dcnv4/modules/dcnv4.py
code served (permissive licence) · get_code("d4124ae005136b98")
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 an instance of its own class, DCNv4: the run constructed the object and did not call it, so no result was computed. 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 math

torch.manual_seed(0)

# The function signature from the code:
# def forward(ctx, input, offset_mask, kernel_h, kernel_w, stride_h, stride_w, 
#             pad_h, pad_w, dilation_h, dilation_w, group, group_channels, 
#             offset_scale, im2col_step, remove_center):

# Create minimal tensors
batch_size = 2
channels = 4
height = 8
width = 8

input_tensor = torch.randn(batch_size, channels, height, width, dtype=torch.float32)

# offset_mask should be shaped for deformable conv operations
# typically [batch, kernel_h * kernel_w * (2 + 1) * group, height, width]
# Using small kernel 3x3 and 1 group
kernel_h = 3
kernel_w = 3
offset_mask_tensor = torch.randn(
    batch_size, 
    kernel_h * kernel_w * 3 * 1,  # 3 for (offset_x, offset_y, mask)
    height, 
    width,
    dtype=torch.float32
)

stride_h = 1
stride_w = 1
pad_h = 1
pad_w = 1
dilation_h = 1
dilation_w = 1
group = 1
group_channels = channels // group
offset_scale = 1.0
im2col_step = 32
remove_center = False

ARGS = {
    'input': input_tensor,
    'offset_mask': offset_mask_tensor,
    'kernel_h': kernel_h,
    'kernel_w': kernel_w,
    'stride_h': stride_h,
    'stride_w': stride_w,
    'pad_h': pad_h,
    'pad_w': pad_w,
    'dilation_h': dilation_h,
    'dilation_w': dilation_w,
    'group': group,
    'group_channels': group_channels,
    'offset_scale': offset_scale,
    'im2col_step': im2col_step,
    'remove_center': remove_center,
}
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 DCNv4 — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DCNv4 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
DCNv4Function Ran
ran on a synthesized input · constructed an object; no result was computed
chenller/mmseg-extension/mmsegextlib/dcnv4/mmsegextlib_dcnv4/modules/dcnv4.py
code served (permissive licence) · get_code("ce961555262652ff")
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 an instance of its own class, DCNv4Function: the run constructed the object and did not call it, so no result was computed. 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 torch.nn.functional as F

torch.manual_seed(0)

# Create minimal tensors for the function
batch_size = 2
height = 56
width = 56
in_channels = 4
out_channels = 16
kernel_h = 3
kernel_w = 3
num_heads = 1

# Input features
input_tensor = torch.randn(batch_size, height, width, in_channels, dtype=torch.float32)

# Weight tensor
weight_tensor = torch.randn(out_channels, kernel_h, kernel_w, in_channels, dtype=torch.float32)

# Offset tensor (deformable convolution offsets)
offset_tensor = torch.randn(batch_size, height, width, num_heads, kernel_h * kernel_w * 2, dtype=torch.float32)

# Mask tensor (for DCNv4, attention mask)
mask_tensor = torch.randn(batch_size, height, width, num_heads, kernel_h * kernel_w, dtype=torch.float32)

# Bias tensor (optional)
bias_tensor = torch.randn(out_channels, dtype=torch.float32)

# Output gradient for backward pass
grad_output = torch.randn(batch_size, height, width, out_channels, dtype=torch.float32)

ARGS = {
    'ctx': None,  # This will be passed by autograd context
    'input': input_tensor,
    'weight': weight_tensor,
    'offset': offset_tensor,
    'mask': mask_tensor,
    'bias': bias_tensor,
    'kernel_h': kernel_h,
    'kernel_w': kernel_w,
    'stride_h': 1,
    'stride_w': 1,
    'pad_h': 1,
    'pad_w': 1,
    'dilation_h': 1,
    'dilation_w': 1,
    'group': 1,
    'group_channels': in_channels,
    'offset_scale': 1.0,
    'im2col_step': 64,
    'remove_center': 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 DCNv4Function — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a DCNv4Function 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
allreduce_hook Ran
ran; our contract draft was wrong, not the code
opengvlab/dcnv4/classification/ddp_hooks.py
pointer only (licence: MIT); the stored copy get_code serves, from gls0425/linvt, is pointer only (licence: NONE) · get_code("bb1a26fe42d7075c")
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 gls0425/linvt, and GitHub detected no licence in that repository when it was checked. 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.

bf16_compress_hook Ran
ran on a synthesized input
opengvlab/dcnv4/classification/ddp_hooks.py
pointer only (licence: MIT); the stored copy get_code serves, from gls0425/linvt, is pointer only (licence: NONE) · get_code("21bd5be56c8f79d0")
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 gls0425/linvt, and GitHub detected no licence in that repository when it was checked. 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.

build_scheduler Ran
ran; our contract draft was wrong, not the code
opengvlab/dcnv4/classification/lr_scheduler.py
pointer only (licence: MIT); the stored copy get_code serves, from leaplabthu/inline, is pointer only (licence: NONE) · get_code("07e8b4b487c0abb6")
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 leaplabthu/inline, and GitHub detected no licence in that repository when it was checked. 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.

create_logger Ran
ran; our contract draft was wrong, not the code
opengvlab/dcnv4/classification/logger.py
pointer only (licence: MIT); the stored copy get_code serves, from ghrua/seqpe, is pointer only (licence: NONE) · get_code("80450600f7f09b0a")
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 ghrua/seqpe, and GitHub detected no licence in that repository when it was checked. 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.

factors Ran
ran, honoured the contract we drafted
chenller/mmseg-extension/mmsegextlib/dcnv4/mmsegextlib_dcnv4/modules/dcnv4.py
code served (permissive licence) · get_code("c9b8eed480765ae8")
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 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 torch

# The function 'factors' takes a single parameter N (an integer)
# and returns a list of all divisors of N.
# Reading the body: it uses N in range(1, N + 1) and N % i
# So N must be an integer.

N = 12

ARGS = {
    "N": N
}
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 int — 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
findspec Ran
ran, honoured the contract we drafted
chenller/mmseg-extension/mmsegextlib/dcnv4/mmsegextlib_dcnv4/modules/dcnv4.py
code served (permissive licence) · get_code("20c4614598af707f")
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 torch
import numpy as np

torch.manual_seed(0)

ARGS = {
    'B': 64,
    'H': 56,
    'W': 56,
    'G': 4,
    'C': 16
}
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 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
findspec Ran
ran, honoured the contract we drafted
opengvlab/dcnv4/DCNv4_op/DCNv4/functions/flash_deform_attn_func.py
code served (permissive licence) · get_code("1c0a5794f6556b80")
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 torch

torch.manual_seed(0)

# Analyzing the function:
# - findspec(B, Q, G, C) takes 4 integer parameters
# - B * Q is used in factors() calculation
# - G and C are multiplied together in the condition check
# - The function performs arithmetic operations only, no method calls
# - All parameters should be integers

ARGS = {
    "B": 4,
    "Q": 3,
    "G": 2,
    "C": 8,
}
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 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
findspec_bwd Ran
ran, honoured the contract we drafted
opengvlab/dcnv4/DCNv4_op/DCNv4/functions/flash_deform_attn_func.py
code served (permissive licence) · get_code("13649542907b646c")
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 torch

torch.manual_seed(0)

ARGS = {
    "B": 4,
    "Q": 8,
    "G": 2,
    "C": 32
}
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 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
fp16_compress_hook Ran
ran; our contract draft was wrong, not the code
opengvlab/dcnv4/classification/ddp_hooks.py
pointer only (licence: MIT); the stored copy get_code serves, from gls0425/linvt, is pointer only (licence: NONE) · get_code("26bfffb625dfe211")
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 gls0425/linvt, and GitHub detected no licence in that repository when it was checked. 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.

find_spec_bwd Not yet run
unverified (harvested, no recorded run)
chenller/mmseg-extension/mmsegextlib/dcnv4/mmsegextlib_dcnv4/modules/dcnv4.py
code served (permissive licence) · get_code("a626f0d745b8c280")
How it ran

no invocation recorded.

find_spec_bwd Not yet run
unverified (harvested, no recorded run)
opengvlab/dcnv4/DCNv4_op/DCNv4/functions/dcnv4_func.py
code served (permissive licence) · get_code("297f5dc3b864f8ca")
How it ran

no invocation recorded.

findspec Not yet run
unverified (harvested, no recorded run)
opengvlab/dcnv4/DCNv4_op/DCNv4/functions/dcnv4_func.py
code served (permissive licence) · get_code("9c57b9e5a08a964c")
How it ran

no invocation recorded.

get_config Not yet run
unverified (harvested, no recorded run)
opengvlab/dcnv4/classification/config.py
pointer only (licence: MIT); the stored copy get_code serves, from gls0425/linvt, is pointer only (licence: NONE) · get_code("e452206cf1a59c19")
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 gls0425/linvt, and GitHub detected no licence in that repository when it was checked. 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. No invocation recorded.

main Not yet run
unverified (harvested, no recorded run)
opengvlab/dcnv4/classification/extract_feature.py
code served (permissive licence) · get_code("67cdd6568078cc51")
How it ran

no invocation recorded.

rgetattr Not yet run
unverified (harvested, no recorded run)
opengvlab/dcnv4/classification/extract_feature.py
code served (permissive licence) · get_code("44925cf3831bfe79")
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 16 distinct functions with an identity, 7 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 1 the same month, 7 later) and 9 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
get_config 51 other papers
25 dated earlier, 1 the same month, 21 later, 4 same year, order not determinable
2103.14030 (2021-03)
Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
2106.03650 (2021-06)
Shuffle Transformer: Rethinking Spatial Shuffle for Vision Transformer
2106.03746 (2021-06)
Efficient Training of Visual Transformers with Small Datasets
and 48 more not named here
create_logger 35 other papers
20 dated earlier, 13 later, 2 same year, order not determinable
2104.13840 (2021-04)
Twins: Revisiting the Design of Spatial Attention in Vision Transformers
2106.03650 (2021-06)
Shuffle Transformer: Rethinking Spatial Shuffle for Vision Transformer
2106.03746 (2021-06)
Efficient Training of Visual Transformers with Small Datasets
and 32 more not named here
build_scheduler 13 other papers
5 dated earlier, 6 later, 2 same year, order not determinable
2103.14030 (2021-03)
Swin Transformer: Hierarchical Vision Transformer using Shifted Windows
2307.06947 (2023-07)
Video-FocalNets: Spatio-Temporal Focal Modulation for Video Action Recognition
2309.01430 (2023-09)
DAT++: Spatially Dynamic Vision Transformer with Deformable Attention
and 10 more not named here
rgetattr 13 other papers
9 dated earlier, 4 later · 1 only as an assembly brick
2203.04036 (2022-03)
StyleHEAT: One-Shot High-Resolution Editable Talking Face Generation via Pre-trained StyleGAN
2209.02952 (2022-09)
BiFuse++: Self-supervised and Efficient Bi-projection Fusion for 360 Depth Estimation
2211.05778 (2022-11)
InternImage: Exploring Large-Scale Vision Foundation Models with Deformable Convolutions
and 10 more not named here
allreduce_hook 2 other papers
2 later
2403.02308 (2024-03)
Vision-RWKV: Efficient and Scalable Visual Perception with RWKV-Like Architectures
2412.05185 (2024-12)
LinVT: Empower Your Image-level Large Language Model to Understand Videos
bf16_compress_hook 2 other papers
2 later
2403.02308 (2024-03)
Vision-RWKV: Efficient and Scalable Visual Perception with RWKV-Like Architectures
2412.05185 (2024-12)
LinVT: Empower Your Image-level Large Language Model to Understand Videos
fp16_compress_hook 2 other papers
2 later
2403.02308 (2024-03)
Vision-RWKV: Efficient and Scalable Visual Perception with RWKV-Like Architectures
2412.05185 (2024-12)
LinVT: Empower Your Image-level Large Language Model to Understand Videos

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

Repositories linked to this paper

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

Abstract

We introduce Deformable Convolution v4 (DCNv4), a highly efficient and effective operator designed for a broad spectrum of vision applications. DCNv4 addresses the limitations of its predecessor, DCNv3, with two key enhancements: 1. removing softmax normalization in spatial aggregation to enhance its dynamic property and expressive power and 2. optimizing memory access to minimize redundant operations for speedup. These improvements result in a significantly faster convergence compared to DCNv3 and a substantial increase in processing speed, with DCNv4 achieving more than three times the forward speed. DCNv4 demonstrates exceptional performance across various tasks, including image classification, instance and semantic segmentation, and notably, image generation. When integrated into generative models like U-Net in the latent diffusion model, DCNv4 outperforms its baseline, underscoring its possibility to enhance generative models. In practical applications, replacing DCNv3 with DCNv4 in the InternImage model to create FlashInternImage results in up to 80% speed increase and further performance improvement without further modifications. The advancements in speed and efficiency of DCNv4, combined with its robust performance across diverse vision tasks, show its potential as a foundational building block for future vision models.

For agents

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

get_harvested_code_for_paper("2401.06197")
get_code_for_paper("2401.06197")
have("2401.06197")

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

curl https://syntology.ai/api/ran/2401.06197.json

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

[![Syntology run record](https://syntology.ai/api/ran/2401.06197.svg)](https://syntology.ai/paper/2401.06197)

Connect an agent — have() is free.