Pith. sign in

REVIEW 4 major objections 6 minor 12 references

Purge-Gate: Backpropagation-Free Test-Time Adaptation for Point Clouds Classification via Token Purging

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that deleting the tokens whose embeddings diverge most from a reference prototype, before they enter attention layers, adapts a point-cloud classifier to test-time corruption with no gradient updates, and reports it…

desk verdict The purge idea is fresh and the paper is worth reading, but the reported accuracy gains are largely a batch-norm reset effect, not token purging alone. read the letter →

arxiv 2509.09785 v1 pith:O4I62ZWS submitted 2025-09-11 cs.CV

Moslem Yazdanpanah , Ali Bahri , Mehrdad Noori , Sahar Dastani , Gustavo Adolfo Vargas Hakim , David Osowiechi , Ismail Ben Ayed , Christian Desrosiers This is my paper · ORCID
classification cs.CV
keywords pointcloudclassificationtest-timeadaptationtokenpurgingbackpropagation-freedomainshiftattentionmechanismtransformersource-free
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that a point-cloud classification transformer can be made resilient to test-time corruption without any gradient updates, by deleting the tokens whose embeddings lie farthest from a reference prototype before they reach the attention layers. The authors argue that corrupted tokens flatten attention weights toward a uniform distribution, so removing them restores the model's ability to aggregate informative features. They report that this 'token purging' outperforms existing backpropagation-free test-time adaptation methods by an average of 10.3 accuracy points on ModelNet40-C, ShapeNet-C, and ScanObjectNN-C, and that a fully source-free variant performs nearly as well. If correct, the result matters because it turns test-time adaptation into a single cheap forward pass, with no optimizer, no fine-tuning, and no source data, which is practical for real-time 3D perception.

What carries the argument

The purge gate is a parameter-free, per-layer filter that ranks tokens by a divergence measure $\Delta(\mathbf{X}^T_j, G)$ against a prototype $G$, discards the top $L_{\mathrm{pg}}$ tokens before they enter self-attention, and keeps the rest. For PG-SP the prototype is the source embedding mean/variance pair and $\Delta$ is the Mahalanobis distance; for PG-SF the prototype is the CLS token mapped through the query projection and $\Delta$ is cosine distance in key space. The mechanism is motivated by two propositions: heavy noise makes per-token embeddings nearly orthogonal after LayerNorm, driving attention scores toward zero and softmax weights toward uniform, so attention stops concentrating on informative tokens. Because the gate is applied before attention, it repairs the input stream rather than updating weights. The only hyperparameter, the purge size $L_{\mathrm{pg}}$, is chosen at test time by running the model for a small set of purge sizes and keeping the output with minimum entropy, and BatchNorm statistics are reset per forward pass.

What would settle it

Run Purge-Gate against a control that removes the same number of tokens at random on a corruption the method was not tuned for; if random removal matches or beats divergence-based purging, the central claim is false. A second test: attach the discriminative signal specifically to tokens with the largest deviation (for example, a rare geometric feature that also shifts statistics); if purging those tokens consistently hurts accuracy, divergence does not identify harmful tokens.

Watch

Extended reading notes

Core claim

On the paper's terms, the central discovery is that domain shift in point cloud transformers can be countered at the token level: compute a per-token divergence from a prototype, delete the most divergent tokens at the input of each attention layer, and the model classifies corrupted point clouds more accurately. Two prototypes are proposed. PG-SP stores source-domain mean and variance statistics, gathered with Welford's online algorithm or inside a modified LayerNorm, and ranks tokens by Mahalanobis distance. PG-SF needs no source data: it uses the CLS token (projected to the query space) as the prototype and ranks tokens by cosine distance in the key space. The paper reports that PG-SP beats prior backpropagation-free TTA methods by an average of 10.3 accuracy points across three corrupted benchmarks, that PG-SF is the strongest source-free variant, and that both run 12.4 times faster and use 5.5 times less memory than the previous no-backprop baseline.

Load-bearing premise

The method rests on the premise that the tokens whose embeddings are farthest from a source or CLS-token prototype are the very tokens whose removal improves accuracy under domain shift; the theory shows heavy noise flattens attention, but it does not prove that the most divergent tokens are the harmful ones, so the effectiveness of the purge is an empirical heuristic.

Editorial extensions

If this is right

  • Test-time adaptation for point cloud transformers reduces to a single forward pass with no parameter updates, so it is deployable in latency- and memory-constrained 3D perception systems.
  • The same purge gate transfers to state-space backbones (PointMamba), lifting mean accuracy on ScanObjectNN-C and ModelNet-C, so the mechanism is not tied to transformer attention itself.
  • Resetting BatchNorm statistics is part of the recipe: on ScanObjectNN-C, PG-SP's accuracy drops from 49.13% to 39.86% without the reset, so normalization statistics carry much of the adaptation signal.
  • An entropy-based selection over a sparse set of purge sizes $\{0,2,4,8,16,32\}$ is enough to approach the accuracy of an exhaustive purge-size search, avoiding per-corruption tuning.
  • The source-free variant PG-SF shows that a single CLS-token prototype can stand in for stored source statistics, removing the need for source data at test time.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the divergence-to-harm link is as general as the paper suggests, the same gate should transfer to 2D vision transformers under common corruptions, where a few background or texture patches play the role of corrupted tokens; the paper does not test this.
  • The supplementary sweeps reveal that for severe background corruption the optimal purge size lies near $L_{\mathrm{pg}}\approx112$, far outside the $\{0,2,4,8,16,32\}$ search window, so an adaptive or severity-aware purge size could extend the reported gains.
  • The theory predicts uniform attention under heavy noise; a direct test would be to measure attention entropy before and after purging and check that the restored non-uniformity tracks accuracy, which the paper does not report.
  • PG-SF's CLS-token prototype could be combined with per-class prototypes to handle partial domain shifts where the global prototype is skewed, an extension the paper leaves open.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper introduces Purge-Gate (PG), a backpropagation-free test-time adaptation method for 3D point cloud classification. PG identifies tokens whose embeddings are most divergent from a source-domain or CLS-token prototype and removes them before they enter attention layers, with the purge size selected at test time by an unsupervised entropy criterion. Two variants are proposed: PG-SP, which uses source statistics, and PG-SF, which is source-free. The method is evaluated on ModelNet40-C, ShapeNet-C, and ScanObjectNN-C, and also on a Mamba backbone, reporting consistent accuracy gains over backpropagation-free baselines, plus large efficiency advantages (12.4x faster, 5.5x lower memory). The paper also provides a theoretical analysis of how noise flattens attention weights and positions the method as a token-level alternative to fine-tuning-based TTA.

Significance. If the central claim holds, token-level purging is a novel and practically attractive mechanism for backpropagation-free test-time adaptation: it requires no parameter updates, is architecture-agnostic (demonstrated on Transformer and Mamba), and shows substantial gains on several corruption benchmarks. The paper includes useful assets: code, reproduction of baseline methods, ablations of the purge-size schedule and BatchNorm reset, and results on real-world ScanObjectNN variants. However, the headline attribution of the gains to token purging is not currently supported by the evidence because the method always couples purging with a BatchNorm-statistics reset, and the only ablation shows that the reset alone accounts for most of the improvement. The paper also openly acknowledges in the supplementary that its fixed purge-size window is far from optimal for severe corruptions, which weakens the claim that entropy-based selection is a reliable test-time hyperparameter mechanism. The significance of the contribution is therefore contingent on an additional control experiment that isolates the marginal effect of purging.

major comments (4)
  1. [4.4 / Fig. 4a, Table 1] The reported gains cannot be attributed to token purging because the method always couples purging with a BatchNorm-statistics reset, and the only provided ablation shows that the reset, not purging, drives most of the improvement. On ScanObjectNN-C, PG-SP with BN reset reaches 49.13% versus 39.86% without reset (+9.27), while the same reset moves BFTT3D from 33.00% to 45.76% (+12.76). Table 1 nevertheless compares PG-SP (which includes the reset) against BFTT3D (which does not include the reset), producing a misleading +16.13 point margin. No experiment reports the Lpg=0 condition with BN reset enabled, so the marginal contribution of token purging itself is never isolated. The authors should report (i) the Lpg=0 + BN-reset condition, (ii) BFTT3D with BN reset in all main tables, and (iii) for both methods, results with and without the reset in the same table.
  2. [6.3 / Fig. 6b, Sec. 4.1] The paper's own supplementary results show that the fixed candidate purge-size set Lpg={0,2,4,8,16,32} used in all experiments is far from optimal for severe corruptions. For Background corruption at severity 7, Fig. 6b reports that the best accuracy reachable within [0,32] is about 16%, while the true optimum near Lpg=112 reaches about 33%. The main text (Sec. 4.4) claims that the entropy-based selection 'achieves a close to perfect accuracy' and validates it on Background and Distortion, but Fig. 6b explicitly states that severe corruption demands far more aggressive purging than the conventional search window can capture. This is a self-acknowledged limitation of the hyperparameter selection mechanism, and it directly affects the reliability of the reported gains on high-severity corruptions. The authors should either make the candidate set adaptive, or clearly state in the main text that the entropy criterion is restricted to the fixed window and does not generalize to severe shifts.
  3. [3.1-3.2, Eqs. (10)-(19)] The theoretical analysis does not establish the core mechanism. The propositions in Sec. 3.1 show that noise-dominant tokens push attention weights toward uniform (Eq. (9)), but they do not imply that (i) the tokens with largest divergence from the prototype are the ones causing the degradation, (ii) removing them restores informative attention, or (iii) the entropy of the classification logits is a valid proxy for accuracy when selecting Lpg. Equations (10)-(11), (15), and (19) implement exactly these heuristics, yet the connection to the propositions is not demonstrated. The paper should explicitly frame the divergence-based selection and the entropy criterion as empirical heuristics, and provide supporting evidence (e.g., a per-token attribution analysis) that the most divergent tokens are the harmful ones under common corruptions. As it stands, the method is presented as derived from the theory, which overstates the theoretical support.
  4. [3.2, Fig. 2, Sec. 4.1] The layer at which purging is applied is ambiguous. The text says tokens are removed 'before they reach the attention layer' and Fig. 2 shows purging at the input of the attention layer, but transformer-based point cloud classifiers contain multiple attention layers. It is not specified whether PG is applied at the first layer only, at every layer, or at a chosen subset. This matters for reproducibility because the source-statistics prototype for PG-SP (Eq. (15)) must be collected from the layer where purging occurs, and the efficiency and accuracy results depend on this choice. The implementation details (Sec. 4.1) do not state the layer placement. Please specify the exact placement and justify it experimentally.
minor comments (6)
  1. [Eq. (13)] In Eq. (13), the expression for σ_i uses µ^S, which is not defined at that point; it should presumably be µ_i, the per-sample mean from Eq. (12). Equation (14) also uses ϕ_i before defining ϕ.
  2. [Eq. (20)] In Eq. (20), L is used both as the purge-size variable and as the name of the candidate set Lpg, which is confusing. A different symbol (e.g., ℓ) should be used for the candidate values.
  3. [3.3.2] The phrase 'reset BatchNorm statistics before each forward pass' is ambiguous: it could mean resetting to stored source statistics or recomputing statistics from the current batch. The following sentence says 'ensuring that it utilizes the current batch statistics', which suggests recomputation, but this should be stated explicitly because it is essential to understanding the method.
  4. [4.3] The text says PG-SP surpasses BFTT3D by +16.13% on ScanObjectNN-C, but Fig. 4a shows that BFTT3D with BN reset reaches 45.76% versus PG-SP's 49.13%, a difference of only 3.37 points. The comparison should be presented with the reset condition clearly noted.
  5. [Tables 1-4] All results are reported from single runs without error bars or variance estimates. For small differences, such as the ShapeNet-C gains of +2.82 and +3.05, and the Mamba ShapeNet-C result of -0.26, it is unclear whether these are statistically meaningful.
  6. [4.2, Table 4] In the paragraph introducing Table 4, 'Table table 4' should be 'Table 4'. Additionally, the description of how ScanObjectNN-C corruptions are generated cites only the ModelNet-C robustness paper; please clarify the specific corruption settings used for ScanObjectNN-C.

Circularity Check

0 steps flagged · score 0.0 of 10

No material circularity: the PG criterion is an unsupervised heuristic validated against external benchmarks, and no load-bearing step is equivalent to its inputs by construction.

full rationale

The central mechanism (Eqs. 10-11, 15, 19) defines token divergence relative to a source-distribution prototype or the pretrained CLS token, and purges the most divergent tokens before attention. This is an operational definition of the heuristic, not a derivation that presupposes the accuracy gain: the gain is measured against external labels on ModelNet40-C, ShapeNet-C, and ScanObjectNN-C. The hyperparameter Lpg is chosen at test time by minimum prediction entropy (Eq. 20), a standard unsupervised TTA criterion, and is not fitted to the classification labels whose accuracy is then reported. The theoretical propositions in Sec. 3.1 (local Lipschitz continuity of LN, near-orthogonality of random directions in high dimension) are self-contained mathematical statements and do not encode the outcome that purging improves accuracy; they motivate but do not logically force the heuristic. Self-citations in the introduction and related work (Yazdanpanah et al. 2022, Osowiechi et al. 2024, Bahri et al. 2024/2025, etc.) are contextual and not load-bearing; the CLS-token premise cites external work (Zou et al. 2024). The BatchNorm-reset ablation in Fig. 4a raises a legitimate attribution concern about how much of the gain is due to purging versus statistic reset, and the supplementary shows the fixed Lpg window misses optimal values on severe corruption; however, these are empirical/confounding issues, not a reduction of the prediction to its inputs by construction. Therefore no circular step is present.

Assumptions & free parameters 1 free parameters · 6 assumptions · 0 invented entities

The method contributes a filtering rule and a hyperparameter-selection heuristic. Its success rests on an unproven correlation between embedding divergence to a prototype and the harmfulness of a token under domain shift, plus the adequacy of six fixed purge sizes. The theoretical propositions are standard results that motivate, but do not establish, the purging rule.

free parameters (1)
  • Candidate purge-size set Lpg = {0, 2, 4, 8, 16, 32}
    Chosen by hand and fixed across all experiments; entropy selection picks a single value. The supplement shows this window is too narrow for severe Background corruption, where the best accuracy in [0,32] is far below the global optimum (38% vs 64% at level 3).
assumptions (6)
  • standard math LayerNorm is locally Lipschitz on compact sets with bounded per-token standard deviation.
    Invoked in Section 3.1, Proposition 1. It is a standard smoothness result, not specific to the paper.
  • standard math Random vectors on a high-dimensional sphere have near-zero expected inner product.
    Invoked in Section 3.1, Proposition 2 via concentration of measure.
  • domain assumption Test-time domain shift can be modeled as additive noise in the embedding space.
    Section 3 states 'we assume that this domain shift can be modeled as an unknown noise function epsilon in the embedding space.' This underpins the entire attention analysis.
  • ad hoc to paper Tokens with the largest divergence from a source or CLS prototype are the most corrupted and should be removed.
    Core heuristic of PG in Section 3.2; not derived from the propositions and only validated empirically on the tested corruptions.
  • ad hoc to paper Output entropy is a valid proxy for classification accuracy when selecting Lpg.
    Used in Section 3.3.1, Eq. (20); the authors validate on a few corruptions (Fig. 5), but the supplement shows the fixed candidate set fails to reach optimal accuracy on severe corruptions.
  • domain assumption The CLS token summarizes source token distribution and absorbs domain information.
    Invoked in Section 3.2.2, citing Zou et al. 2024. Basis for the source-free prototype G in Eq. (18).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Purge-Gate: Backpropagation-Free Test-Time Adaptation for Point Clouds Classification via Token Purging." pith.science (2026). https://pith.science/paper/O4I62ZWS

@misc{pith2026250909785,
  author       = {Pith},
  title        = {Pith review of: Purge-Gate: Backpropagation-Free Test-Time Adaptation for Point Clouds Classification via Token Purging},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O4I62ZWS}},
  note         = {Machine review of arXiv:2509.09785}
}
read the original abstract

Test-time adaptation (TTA) is crucial for mitigating performance degradation caused by distribution shifts in 3D point cloud classification. In this work, we introduce Token Purging (PG), a novel backpropagation-free approach that removes tokens highly affected by domain shifts before they reach attention layers. Unlike existing TTA methods, PG operates at the token level, ensuring robust adaptation without iterative updates. We propose two variants: PG-SP, which leverages source statistics, and PG-SF, a fully source-free version relying on CLS-token-driven adaptation. Extensive evaluations on ModelNet40-C, ShapeNet-C, and ScanObjectNN-C demonstrate that PG-SP achieves an average of +10.3\% higher accuracy than state-of-the-art backpropagation-free methods, while PG-SF sets new benchmarks for source-free adaptation. Moreover, PG is 12.4 times faster and 5.5 times more memory efficient than our baseline, making it suitable for real-world deployment. Code is available at \hyperlink{https://github.com/MosyMosy/Purge-Gate}{https://github.com/MosyMosy/Purge-Gate}

Figures

Figures reproduced from arXiv: 2509.09785 by the authors.

Figure 1
Figure 1. Token-level domain shift visualization. From left to right: clean point cloud, corrupted point cloud, tokens [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The overall procedure of PG algorithm at the input of the attention layer. Here, t denotes the token index and Lt is the total number of tokens for each point cloud. The center point ct is chosen via a furthest-point strategy, and KNN (ct) selects the k-nearest neighbors of ct. This tokenization preserves the local neighborhood structure in the 3D point cloud, which is beneficial for subsequent attention-based layer… view at source ↗
Figure 3
Figure 3. Effect of Batch size on ACC performance of our [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Comparison of our PG-SP method and the baseline BFTT3D on the ScanObjectNN-C and ModelNet-C datasets. (a) Effect of BatchNorm reset on accuracy. (b) GPU time and memory usage with batch size 32. 024 8 16 32 48 64 80 96 112 128 Purge Size 10 20 30 40 50 Top1 Acc Top1 AC…
Figure 5
Figure 5. Figure 5: Effect of purge size on ACC under Background (left) and Distortion (right) corruptions from the ScanObjectNN-C dataset. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Effect of purge size on ACC of Background corruption at different severity levels (3 and 7) from the ScanObjectNN-C dataset. 15 [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 2 canonical work pages

  1. [4]

    Evaluating prediction-time batch normalization for robustness under covariate shift.arXiv preprint arXiv:2006.10963,

    Zachary Nado, Shreyas Padhy, D Sculley, Alexander D’Amour, Balaji Lakshminarayanan, and Jasper Snoek. Evaluating prediction-time batch normalization for robustness under covariate shift.arXiv preprint arXiv:2006.10963,

    arXiv 2006

  2. [6]

    Continual test-time domain adaptation via dynamic sample selection

    Yanshuo Wang, Jie Hong, Ali Cheraghian, Shafin Rahman, David Ahmedt-Aristizabal, Lars Petersson, and Mehrtash Harandi. Continual test-time domain adaptation via dynamic sample selection. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 1701–1710, 2024b. Ali Bahri, Moslem Yazdanpanah, Mehrdad Noori, Sahar Dastani, M...

    arXiv

  3. [8]

    SMART–PC: Skeletal model adaptation for ro- bust test–time training in point clouds.Proceedings of the 42 nd International Conference on Machine Learning (ICML 2025), 2025a

    Ali Bahri, Moslem Yazdanpanah, Sahar Dastani, Mehrdad Noori, Gustavo Adolfo Vargas Hakim, David Osowiechi, Farzad Beizaee, Ismail Ben Ayed, and Christian Desrosiers. SMART–PC: Skeletal model adaptation for ro- bust test–time training in point clouds.Proceedings of the 42 nd International Conference on Machine Learning (ICML 2025), 2025a. Inkyu Shin, Yi-Hs...

    work page 2025

  4. [9]

    Ppt: Token pruning and pooling for efficient vision transformers.arXiv preprint arXiv:2310.01812,

    Xinjian Wu, Fanhu Zeng, Xiudong Wang, and Xinghao Chen. Ppt: Token pruning and pooling for efficient vision transformers.arXiv preprint arXiv:2310.01812,

    arXiv

  5. [11]

    Shapenet: An information-rich 3d model repository.arXiv preprint arXiv:1512.03012,

    Angel X Chang, Thomas Funkhouser, Leonidas Guibas, Pat Hanrahan, Qixing Huang, Zimo Li, Silvio Savarese, Manolis Savva, Shuran Song, Hao Su, et al. Shapenet: An information-rich 3d model repository.arXiv preprint arXiv:1512.03012,

    arXiv

  6. [2015]

    Benchmarking robustness of 3d point cloud recognition against common corruptions.arXiv preprint arXiv:2201.12296,

    Jiachen Sun, Qingzhao Zhang, Bhavya Kailkhura, Zhiding Yu, Chaowei Xiao, and Z Morley Mao. Benchmarking robustness of 3d point cloud recognition against common corruptions.arXiv preprint arXiv:2201.12296,

    arXiv

  7. [2019]

    6 Supplementary Material 6.1 Effect of Batch Size

    12 Purge-Gate: Backpropagation-Free Test-Time Adaptation for Point Clouds Classification via Token Purging Variant Source Only PG-SP (Ours) ∆ OBJ-BG 74.1875.56 +1.38 PB-T50-RS 56.7761.07 +4.30 Table 5: Top-1 classification accuracy (%) on real-world ScanObjectNN variants without synthetic corruptions. 6 Supplementary Material 6.1 Effect of Batch Size. Fig...

    work page 2000

  8. [2020]

    Tent: Fully test-time adaptation by entropy minimization.arXiv preprint arXiv:2006.10726,

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. Tent: Fully test-time adaptation by entropy minimization.arXiv preprint arXiv:2006.10726,

    arXiv 2006

Show all 12 references
  1. [2021]

    TTN: A domain-shift aware batch normalization in test-time adaptation.arXiv preprint arXiv:2302.05155,

    Hyesu Lim, Byeonggeun Kim, Jaegul Choo, and Sungha Choi. TTN: A domain-shift aware batch normalization in test-time adaptation.arXiv preprint arXiv:2302.05155,

    arXiv

  2. [2022]

    Mamba: Linear-time sequence modeling with selective state spaces.arXiv preprint arXiv:2312.00752,

    Albert Gu and Tri Dao. Mamba: Linear-time sequence modeling with selective state spaces.arXiv preprint arXiv:2312.00752,

    arXiv

  3. [2023]

    GeoMask3D: Geometrically informed mask selection for self-supervised point cloud learning in 3D.arXiv preprint arXiv:2405.12419, 2024a

    Ali Bahri, Moslem Yazdanpanah, Mehrdad Noori, Milad Cheraghalikhani, Gustavo Adolfo Vargas Hakim, David Osowiechi, Farzad Beizaee, Ismail Ben Ayed, and Christian Desrosiers. GeoMask3D: Geometrically informed mask selection for self-supervised point cloud learning in 3D.arXiv p...

    arXiv

  4. [2025]

    Test-time adaptation of vision-language models for open-vocabulary semantic segmentation.arXiv preprint arXiv:2505.21844,

    Mehrdad Noori, David Osowiechi, Gustavo Adolfo Vargas Hakim, Ali Bahri, Moslem Yazdanpanah, Sahar Dastani, Farzad Beizaee, Ismail Ben Ayed, and Christian Desrosiers. Test-time adaptation of vision-language models for open-vocabulary semantic segmentation.arXiv preprint arXiv:2...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.