SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2103.07941 · CVPR · 2021

Modular Interactive Video Object Segmentation: Interaction-to-Mask, Propagation and Difference-Aware Fusion

Ho Cheng, Yu-Wing Tai

arXiv · PDF · Open in the Atlas

Code that ran

10 of 18 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (3 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 5 have no contract verdict; 5 failures of our instrument; 8 no recorded run), measured 2026-11-21; 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/2103.07941.json · badge: /api/ran/2103.07941.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
Vujas-Eteph/CiVOS community
Papers with Code archive
7 ran · 3 no recorded run
hkchengrex/MiVOS marked official in Papers with Code
Papers with Code archive
2 ran · 2 no recorded run
hkchengrex/Mask-Propagation community
Papers with Code archive
1 ran · 1 no recorded run
limingxing00/rde-vos-cvpr2022 community
Papers with Code archive
0 ran · 2 no recorded run
FunctionStatusWhere it lives
AttentionMemory Ran
ran on a synthesized input · constructed an object; no result was computed
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("9a8a4b812337da11")
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, AttentionMemory: the run constructed the object and did not call it, so no result was computed. 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 math
import torch.nn as nn
import torch.nn.functional as F

torch.manual_seed(0)

class AttentionMemory(nn.Module):

    def __init__(self, k):
        super().__init__()
        self.k = k

    def forward(self, mk, qk):
        """
        T=1 only. Only needs to obtain W
        """
        B, CK, _, H, W = mk.shape
        mk = mk.view(B, CK, H * W)
        mk = torch.transpose(mk, 1, 2)
        qk = qk.view(1, CK, H * W).expand(B, -1, -1) / math.sqrt(CK)
        affinity = torch.bmm(mk, qk)
        affinity = F.softmax(affinity, dim=1)
        return affinity

# Create attention memory module
attention_memory = AttentionMemory(k=4)

# Create mk tensor: shape (B, CK, T, H, W) where T=1
# B=2, CK=4, T=1, H=3, W=5
mk = torch.randn(2, 4, 1, 3, 5)

# Create qk tensor: shape (1, CK, HW) or can be (1, CK, 1)
# CK=4 to match mk
qk = torch.randn(1, 4, 15)

ARGS = {
    "k": 4,
}
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 AttentionMemory — 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 AttentionMemory 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
AttentionMemory Ran
ran on a synthesized input · constructed an object; no result was computed
hkchengrex/MiVOS/model/propagation/prop_net.py
code served (permissive licence) · get_code("e72b80185fd5e21a")
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, AttentionMemory: the run constructed the object and did not call it, so no result was computed. 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)

# AttentionMemory takes parameter 'k' in __init__
# Looking at the forward method:
# - mk has shape (B, CK, T, H, W) where T=1
# - qk has shape (1, CK, H*W)
# - The method doesn't actually use self.k, but we need to provide it

# Create small tensors for testing
B = 2  # batch size
CK = 4  # channel/key dimension
T = 1  # time dimension (only T=1 supported)
H = 3  # height
W = 5  # width

# mk shape: (B, CK, T, H, W)
mk = torch.randn(B, CK, T, H, W)

# qk shape: (1, CK, H*W)
qk = torch.randn(1, CK, H * W)

# k parameter for AttentionMemory
k = 4

ARGS = {
    'k': k,
}
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 AttentionMemory — 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 AttentionMemory 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
EvalMemoryReader Ran
ran on a synthesized input · constructed an object; no result was computed
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("d7a09a5e37f31c1e")
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, EvalMemoryReader: 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 math
import torch.nn as nn
import torch.nn.functional as F

torch.manual_seed(0)

# Create small tensors with distinct dimensions
B = 2  # batch size
CK = 4  # key channels
CV = 3  # value channels
T = 2  # time steps
H = 5  # height
W = 6  # width

# Memory key: (B, CK, T, H, W)
mk = torch.randn(B, CK, T, H, W)

# Memory value: (B, CV, T, H, W)
mv = torch.randn(B, CV, T, H, W)

# Query key: (1, CK, H, W)
qk = torch.randn(1, CK, H, W)

# Parameters for EvalMemoryReader
top_k = 3
km = 5.0  # sigma for gaussian (can be None or a float value)

ARGS = {
    "top_k": top_k,
    "km": km
}
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 EvalMemoryReader — 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 EvalMemoryReader 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
FusionNet Ran
ran on a synthesized input
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("3d4477934fb35965")
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 4, 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

INIT_ARGS = {}

ARGS = {
    "im": torch.randn(2, 3, 8, 6),
    "seg1": torch.randn(2, 1, 8, 6),
    "seg2": torch.randn(2, 1, 8, 6),
    "attn": torch.randn(2, 3, 8, 6),
    "time": torch.randn(2, 1)
}
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 4, 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)
MemoryReader Ran
ran on a synthesized input
hkchengrex/Mask-Propagation/model/network.py
code served (permissive licence) · get_code("d9ba63b1718e1e05")
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 4, 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 math
import torch
import torch.nn as nn
import torch.nn.functional as F

class MemoryReader(nn.Module):

    def __init__(self):
        super().__init__()

    def forward(self, mk, mv, qk, qv):
        B, CK, T, H, W = mk.shape
        _, CV, _, _, _ = mv.shape
        mi = mk.view(B, CK, T * H * W)
        mi = torch.transpose(mi, 1, 2)
        qi = qk.view(B, CK, H * W) / math.sqrt(CK)
        affinity = torch.bmm(mi, qi)
        affinity = F.softmax(affinity, dim=1)
        mv = mv.view(B, CV, T * H * W)
        mem = torch.bmm(mv, affinity)
        mem = mem.view(B, CV, H, W)
        mem_out = torch.cat([mem, qv], dim=1)
        return mem_out


INIT_ARGS = {}

ARGS = {
    'mk': torch.randn(2, 4, 3, 5, 6),
    'mv': torch.randn(2, 7, 3, 5, 6),
    'qk': torch.randn(2, 4, 5, 6),
    'qv': torch.randn(2, 8, 5, 6),
}
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 4, 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)
aggregate_wbg Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("d763f9348cf73778")
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, 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
import torch.nn.functional as F

torch.manual_seed(0)

# Create a small probability tensor with shape (k, 1, h, w)
# k=2 (batch), 1 channel, h=4, w=3
prob = torch.rand(2, 1, 4, 3)
# Ensure values are in valid probability range [0, 1]
prob = torch.clamp(prob, 0.1, 0.9)

ARGS = {
    "prob": prob,
    "keep_bg": False,
    "hard": 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 4, dtype family float, shape (2, 1, 4, 3) — shape (2, 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_gaussian Ran
ran; our contract draft was wrong, not the code
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("de847b7a64641452")
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; no output shape is stated. 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 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)

# The function expects:
# - y_idx: a tensor that will be transposed (0,1) and used in shape (1, N)
# - x_idx: a tensor that will be transposed (0,1) and used in shape (1, N)
# - height: an integer for image height
# - width: an integer for image width
# - sigma: optional float (default 7)

# Create small tensors with distinct dimensions
# y_idx and x_idx need to be 2D tensors that will be transposed
# After transpose(0,1), they should have shape (N, 1) where N matches num_points
batch_size = 2
num_points = 4

y_idx = torch.randint(0, 5, (1, batch_size)).float()
x_idx = torch.randint(0, 5, (1, batch_size)).float()

height = 3
width = 4
sigma = 7

ARGS = {
    "y_idx": y_idx,
    "x_idx": x_idx,
    "height": height,
    "width": width,
    "sigma": sigma
}
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 — no output shape is stated: observe_signature.py could not observe it (call:AssertionError)
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
pad_divide_by Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("29ef75051beaf38a")
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.functional as F

torch.manual_seed(0)

# Create a small tensor with shape (batch, channels, height, width)
in_img = torch.randn(2, 3, 4, 5)

# Divisor value
d = 3

# Optional in_size parameter (height, width)
in_size = None

ARGS = {
    "in_img": in_img,
    "d": d,
    "in_size": in_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
softmax_w_g_top Ran
ran; our contract draft was wrong, not the code
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("7d8c831d1808fdcc")
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)

# Create a small batch of logits: shape (batch_size=2, num_classes=4)
x = torch.randn(2, 4)

# Create a small gaussian weighting tensor with same shape
gauss = torch.ones(2, 4) * 0.5

# Set top-k parameter
top = 2

ARGS = {
    "x": x,
    "top": top,
    "gauss": gauss,
}
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
softmax_w_g_top Ran
ran; our contract draft was wrong, not the code
hkchengrex/MiVOS/model/propagation/prop_net.py
code served (permissive licence) · get_code("ada9a3eef6b445bd")
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)

# Create a small input tensor (batch_size=2, num_features=4)
x = torch.randn(2, 4, requires_grad=False)

# Create a gauss tensor with the same shape as x for element-wise multiplication
gauss = torch.ones(2, 4)

# top parameter for selecting top-k elements
top = 2

ARGS = {
    "x": x,
    "top": top,
    "gauss": gauss,
}
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
Decoder Not yet run
unverified (harvested, no recorded run)
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("3c2bb88e4d9e2e6f")
How it ran

no invocation recorded.

EvalMemoryReader Not yet run
unverified (harvested, no recorded run)
hkchengrex/MiVOS/model/propagation/prop_net.py
code served (permissive licence) · get_code("9958d67d9d027e4f")
How it ran

no invocation recorded.

InferenceCore Not yet run
unverified (harvested, no recorded run)
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("1f74746f70cfb8ea")
How it ran

no invocation recorded.

MemoryReader Not yet run
unverified (harvested, no recorded run)
limingxing00/rde-vos-cvpr2022/model/network.py
code served (permissive licence) · get_code("8cb186830eed0c02")
How it ran

no invocation recorded.

PropagationNetwork Not yet run
unverified (harvested, no recorded run)
Vujas-Eteph/CiVOS/inference_core.py
code served (permissive licence) · get_code("2d76453bfdfe5c8f")
How it ran

no invocation recorded.

PropagationNetwork Not yet run
unverified (harvested, no recorded run)
hkchengrex/Mask-Propagation/model/network.py
code served (permissive licence) · get_code("046904ddbd853941")
How it ran

no invocation recorded.

PropagationNetwork Not yet run
unverified (harvested, no recorded run)
hkchengrex/MiVOS/model/propagation/prop_net.py
code served (permissive licence) · get_code("b9081800a0d70e91")
How it ran

no invocation recorded.

RDE_VOS Not yet run
unverified (harvested, no recorded run)
limingxing00/rde-vos-cvpr2022/model/network.py
code served (permissive licence) · get_code("afa77ff620b598c6")
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 17 distinct functions with an identity, 3 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 3 later) and 14 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

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

FunctionAlso present inEarliest-datedOthers named
Decoder
brick of this paper's assembly: helper, model
2 other papers
2 later · 2 only as an assembly brick
2207.07922 (2022-07)
Learning Quality-aware Dynamic Memory for Video Object Segmentation
2601.00988 (2026-01)
Few-Shot Video Object Segmentation in X-Ray Angiography Using Local Matching and Spatio-Temporal Consistency Loss
MemoryReader
brick of this paper's assembly: model
1 other paper
1 later · 1 only as an assembly brick
2207.07922 (2022-07)
Learning Quality-aware Dynamic Memory for Video Object Segmentation
pad_divide_by
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2208.01924 (2022-08)
Per-Clip Video Object Segmentation

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 2026-11-21.

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 present Modular interactive VOS (MiVOS) framework which decouples interaction-to-mask and mask propagation, allowing for higher generalizability and better performance. Trained separately, the interaction module converts user interactions to an object mask, which is then temporally propagated by our propagation module using a novel top-k filtering strategy in reading the space-time memory. To effectively take the user's intent into account, a novel difference-aware module is proposed to learn how to properly fuse the masks before and after each interaction, which are aligned with the target frames by employing the space-time memory. We evaluate our method both qualitatively and quantitatively with different forms of user interactions (e.g., scribbles, clicks) on DAVIS to show that our method outperforms current state-of-the-art algorithms while requiring fewer frame interactions, with the additional advantage in generalizing to different types of user interactions. We contribute a large-scale synthetic VOS dataset with pixel-accurate segmentation of 4.8M frames to accompany our source codes to facilitate future research.

For agents

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

get_harvested_code_for_paper("2103.07941")
get_code_for_paper("2103.07941")
have("2103.07941")

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

curl https://syntology.ai/api/ran/2103.07941.json

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

[![Syntology run record](https://syntology.ai/api/ran/2103.07941.svg)](https://syntology.ai/paper/2103.07941)

Connect an agent — have() is free.