Pith. sign in

REVIEW 4 major objections 4 minor 48 references

Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Finer-CAM proposes that class activation maps should explain the difference between a target class and a visually similar class, not the target class alone.

desk verdict Clean idea, weak evaluation: Finer-CAM's logit-difference framing is a nice practical trick, but the headline metric is aligned with its objective, so independent evidence is still needed. read the letter →

arxiv 2501.11309 v2 pith:PUG3EPOQ submitted 2025-01-20 cs.CV cs.AI

Ziheng Zhang , Jianyang Gu , Arpita Chowdhury , Zheda Mai , David Carlyn , Tanya Berger-Wolf , Yu Su , Wei-Lun Chao This is my paper · ORCID
classification cs.CVcs.AI
keywords classactivationmapfine-grainedvisualexplanationsaliencylogitdifferencerelativeconfidencedropzero-shotlocalizationinterpretabilityexplainableAI
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

Finer-CAM claims that class activation maps fail on fine-grained tasks not because of how they combine feature maps but because of what they are asked to explain: the target class's prediction score alone. The paper proposes instead to explain the difference between the target score and the score of a visually similar class (the logit difference), so that features shared with the lookalike are suppressed and the features that genuinely separate the classes are highlighted. If correct, this means any CAM variant can gain fine-grained precision simply by changing the explanation target, with no extra computation or training. The authors demonstrate the mechanism on gradient-based and score-based CAMs, and extend it to CLIP-style text prompts for localizing specific concepts in an image.

What carries the argument

The machinery is the logit-difference target $y_c - \gamma y_d$ in Eq. (5), where $y_c$ and $y_d$ are the raw class scores for the target and a visually similar class, and $\gamma$ is the comparison strength. For gradient-based CAMs, differentiation linearity turns this into a per-channel weight subtraction $\alpha^{c,d}_k = \alpha^c_k - \gamma \alpha^d_k$ (Eq. 7) before the ReLU, which differs from subtracting saliency maps after the fact. This object carries the argument because it converts a class-conditional explanation into a contrastive one, and the adjustable $\gamma$ provides a continuous dial between coarse object-level activation and fine discriminative detail.

What would settle it

Take a target class and pair it with its most similar class and with its least similar class, then compute Finer-CAM maps for both pairs. If the least-similar pairing matches or beats the most-similar pairing on relative confidence drop at 5% masking, then the comparison of similar classes is not what drives the reported gain.

Watch

Extended reading notes

Core claim

The central discovery is that the explanation target, not the explanation machinery, is what limits CAM in fine-grained settings. Formally, Finer-CAM replaces the target logit $y_c$ with $y_c - \gamma y_d$, where $d$ is a visually similar class and $\gamma$ controls comparison strength, and computes CAM weights from $\partial(y_c - \gamma y_d)/\partial A^k_{ij}$. Because the gradient is linear, the resulting weight for channel $k$ is $\alpha^{c,d}_k = \alpha^c_k - \gamma \alpha^d_k$, so the map positively weights features that raise the target logit without also raising the lookalike's logit. The paper shows this yields saliency maps that, when their top 5% of pixels are masked, drop the target class's confidence relative to the similar class's confidence more than baseline CAMs do, and that the same comparison can be applied to score-based CAMs and to zero-shot text-prompt localization.

Load-bearing premise

The load-bearing premise is that the difference between two class scores, $y_c - y_d$, faithfully points to the visual features that distinguish the classes, so that gradients of that difference land on genuinely discriminative image regions.

Editorial extensions

If this is right

  • Any gradient-based CAM (Grad-CAM, Layer-CAM, and similar) can be turned into a finer-grained explainer by substituting the logit-difference derivative, without retraining or extra forward passes.
  • Score-based CAMs get the same recipe via Eq. (9): weight each feature map by how much it raises the target score relative to the reference score.
  • In zero-shot multi-modal models, comparing a concept prompt against a base-category prompt localizes that concept more accurately than prompting the concept alone.
  • The relative confidence drop metric (Eq. 10) tracks the target-vs-lookalike confidence gap rather than target confidence alone, exposing differences that standard deletion curves miss.
  • Aggregating comparisons over the top few similar classes produces a saliency map that captures multiple discriminative traits of the target class.

Reading between the lines

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

  • The same comparison trick could work in other explanation families, such as perturbation- or attention-based methods, wherever a scalar class score is available.
  • Because the reference class is a free choice, the method doubles as a way of asking a model what would make an image look like class $d$ instead of class $c$, which could support interactive debugging of confusions.
  • The paper's failure cases suggest the improvement depends on choosing references with genuinely similar visual features; an adversarial test with least-similar references would show how much of the gain is due to the similarity selection.
  • The multi-modal extension offers a concrete audit tool: where classifier-driven and attribute-driven Finer-CAM maps disagree, either the classifier is using non-attribute cues or the attribute annotation is incomplete.
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 / 4 minor

Summary. The paper proposes Finer-CAM, a modification of CAM-based explanation methods that replaces the explanation target, the target class logit y_c, with the logit difference y_c - γ y_d, where d is a visually similar class. The authors argue that this suppresses features shared with similar classes and highlights discriminative details. They derive the method for gradient-based and score-based CAM, extend it to CLIP-based multi-modal models, and evaluate on fine-grained datasets using deletion AUC, a new relative confidence drop metric, and pointing-game localization.

Significance. If the central claim is correct, the method is valuable: it is a simple, parameter-light modification compatible with existing CAM methods and could improve fine-grained interpretability. The derivation in Eqs. (5)-(7) is mathematically sound, and the paper includes a useful multi-modal extension and qualitative consistency analysis. However, the empirical evidence is not yet convincing because the main quantitative metric is aligned with the method's objective and is used for hyperparameter selection.

major comments (4)
  1. [§3.2, Eq. (5)] The load-bearing premise that gradients of y_c - γ y_d localize visually discriminative features is asserted rather than demonstrated. The paper's own failure cases (Fig. 10) concede that Finer-CAM degenerates when the classifier is wrong or when logit similarity does not reflect visual similarity, which indicates the premise is not guaranteed. To support the central claim, the authors should validate the correspondence between logit differences and human-annotated distinguishing attributes (e.g., using CUB part/attribute annotations) or provide a controlled experiment.
  2. [§3.4, Eq. (10), Table 1] The relative confidence drop metric is defined as (p_c - p_c*) - (p_d - p_d*), which directly measures the quantity that Finer-CAM is designed to maximize: the gap between target and reference logits. Because γ and the number of aggregated references are tuned on this metric on the same datasets (Fig. 6b), the reported RD improvements are expected by construction and do not constitute independent evidence of finer localization. The authors should report an evaluation metric that is not aligned with the training objective, such as insertion/deletion AUC or agreement with fine-grained part annotations.
  3. [Tables 1, 4, and 5] Standard deletion AUC shows negligible or inconsistent differences between Finer-CAM and baselines (e.g., Birds-525 Grad-CAM 0.079 vs. +Finer 0.076; CUB 0.024 vs. 0.024; Cars 0.024 vs. 0.024). Since deletion AUC is the standard faithfulness metric and is not aligned with Finer-CAM's objective, the lack of improvement is a serious concern for the claim that Finer-CAM yields more faithful explanations. The paper should explain this discrepancy or provide a direct, non-aligned measure of discriminative-detail localization.
  4. [§4.2, Localization] The energy-based pointing game only evaluates whether activations fall within the object bounding box, which cannot distinguish between coarse object-level localization and localization of the specific fine-grained details that the method claims to identify. The authors should use part-level or attribute-level annotations (available for CUB-200) to show that Finer-CAM highlights the actual distinguishing regions rather than just the object.
minor comments (4)
  1. [§3.2, Eq. (9)] The baseline term for the reference class appears to be missing; the expression should likely be [f(x◦H_k)c - f(xb)c] - γ[f(x◦H_k)d - f(xb)d] or an explicit assumption that the baseline terms cancel.
  2. [Supplementary B.3] The text uses 'Fine-CAM' where 'Finer-CAM' is intended.
  3. [Figure 6b] The label 'Aggreagte' is a typo for 'Aggregate'.
  4. [§3.4] The paper states that 'the top 5% of activated pixels' are masked but does not specify how the threshold is applied across images of different sizes; please clarify the masking procedure.

Circularity Check

2 steps flagged · score 6.0 of 10

Finer-CAM's headline quantitative claim is evaluated with a relative-confidence-drop metric that is the same objective the method optimizes by construction, and the comparison strength is tuned on that metric; independent deletion AUC shows no advantage.

  1. self definitional [Sec. 3.2 Eq. (5), Sec. 3.4 Eq. (10), Abstract]
    "We propose to additionally involve similar classes and calculate the gradients of the logit difference: α_c,d_k = 1/Z Σ_i Σ_j ∂(y_c − γ × y_d)/∂A_ij_k ... RD = (p_c − p_c⋆) − (p_d − p_d⋆). Larger drops mean that masking the top pixels effectively reduces the confidence in predicting the target class over the reference, indicating more discriminative saliency maps."

    Finer-CAM constructs its saliency map from gradients of y_c − γ·y_d, so the top-activated pixels are, by construction, those whose removal should reduce the target logit relative to the reference logit. The evaluation metric RD is exactly the drop in (p_c − p_d) after masking those pixels. The reported RD improvement is therefore a restatement of the method's objective in confidence space, not an independent confirmation that the highlighted regions are the visually distinguishing details. Eq. (7) makes the construction explicit: α_c,d_k = α_c_k − γ·α_d_k, so the map is the difference of the two class-support maps.

  2. fitted input called prediction [Sec. 4.3 Fig. 6(b), Sec. 4.2 Tab. 1]
    "As the comparison strength increases, the relative drop is boosted at a fast pace and reaches the peak when γ = 0.8. Further strengthening the comparison leads to slight performance drops. ... Based on both metrics, we adopt the strength γ = 0.6 and the aggregation of the top 3 similar classes as our final design."

    The free parameter γ and the number of aggregated similar classes are selected by maximizing the same relative-confidence-drop metric on which the headline improvements in Tab. 1 are reported. The RD gains are thus partly the result of tuning the method's hyperparameters on the evaluation metric itself; they are expected under the selected configuration rather than an independent, out-of-sample confirmation of the method's superiority.

full rationale

The algebraic derivation of Finer-CAM, Eqs. (5)–(7), is not circular: replacing the explanation target y_c with y_c − γ·y_d and applying gradient linearity is a legitimate construction. No load-bearing self-citation or imported uniqueness theorem appears; the citations to the authors' prior interpretable-transformer work are not used to justify the method's core step. The circularity concern is in the evaluation. The paper's central quantitative claim, 'masking out the top 5% of activated pixels by Finer-CAM results in a larger relative confidence drop compared to baselines,' is measured with RD (Eq. 10), which is the confidence-space analogue of the logit difference that Finer-CAM is built to maximize. Additionally, the hyperparameters γ and aggregation choice are tuned on this same metric, further aligning the reported improvements with the method's objective. The paper itself concedes that standard deletion AUC shows only negligible differences ('there only exist negligible differences between baselines and Finer-CAM'), and the localization gains are bounding-box pointing games that cannot verify that the activated regions are the specific fine discriminative details. These factors make the headline quantitative support partially self-definitional, although the method retains independent qualitative content in its multi-modal extension and visualizations. Score 6 reflects partial circularity: the central comparison metric is aligned with the method's objective by construction, while the derivation itself is sound.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The method introduces no new physical or architectural entities. The main free parameter is the comparison strength γ, tuned on the headline metric. The key assumptions are that logit differences faithfully encode visual discriminative traits and that the proposed RD metric validly measures explanation quality.

free parameters (2)
  • comparison strength γ = 0.6
    Controls how strongly features shared with the reference class are suppressed. Selected by maximizing RD@0.05 on Birds-525 (Fig. 6b); the same dataset is then used in the main results, and no separate validation split is mentioned.
  • number of compared reference classes T (aggregation top-k) = 3
    Aggregation over the top 3 similar classes is the default design (Sec. 4.1); Tab. 2 suggests top-3 aggregation gives the best RD among the tested options, but no sensitivity analysis across datasets is provided.
assumptions (4)
  • standard math For a network with global average pooling followed by a linear layer, the Grad-CAM weight α_k equals the linear classifier weight w_k up to a constant (Eq. 4).
    Invoked in Sec. 3.2 to justify that CAM weights reflect classifier weights; this is a standard result in the CAM literature.
  • domain assumption The logit difference y_c - γ y_d is a differentiable signal whose gradients highlight features that distinguish class c from class d.
    Core mechanism of Finer-CAM (Eq. 5). The paper does not prove that gradients of logit difference correspond to human-perceptible discriminative traits; Fig. 10 shows failure cases when the classifier is wrong or classes are not visually similar.
  • domain assumption Similarity between linear classifier weights reflects shared visual features between classes.
    Used in Fig. 1 and to select reference classes; the paper measures average weight similarities but does not establish a general correspondence between weight similarity and visual similarity.
  • ad hoc to paper The relative confidence drop metric (Eq. 10) is a valid measure of fine-grained explanation quality.
    This metric is introduced in the paper and is not an established benchmark; it directly measures the quantity Finer-CAM is designed to maximize, so its use as the primary evidence carries an element of circularity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation." pith.science (2026). https://pith.science/paper/PUG3EPOQ

@misc{pith2026250111309,
  author       = {Pith},
  title        = {Pith review of: Finer-CAM: Spotting the Difference Reveals Finer Details for Visual Explanation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PUG3EPOQ}},
  note         = {Machine review of arXiv:2501.11309}
}
read the original abstract

Class activation map (CAM) has been widely used to highlight image regions that contribute to class predictions. Despite its simplicity and computational efficiency, CAM often struggles to identify discriminative regions that distinguish visually similar fine-grained classes. Prior efforts address this limitation by introducing more sophisticated explanation processes, but at the cost of extra complexity. In this paper, we propose Finer-CAM, a method that retains CAM's efficiency while achieving precise localization of discriminative regions. Our key insight is that the deficiency of CAM lies not in "how" it explains, but in "what" it explains. Specifically, previous methods attempt to identify all cues contributing to the target class's logit value, which inadvertently also activates regions predictive of visually similar classes. By explicitly comparing the target class with similar classes and spotting their differences, Finer-CAM suppresses features shared with other classes and emphasizes the unique, discriminative details of the target class. Finer-CAM is easy to implement, compatible with various CAM methods, and can be extended to multi-modal models for accurate localization of specific concepts. Additionally, Finer-CAM allows adjustable comparison strength, enabling users to selectively highlight coarse object contours or fine discriminative details. Quantitatively, we show that masking out the top 5% of activated pixels by Finer-CAM results in a larger relative confidence drop compared to baselines. The source code and demo are available at https://github.com/Imageomics/Finer-CAM.

Figures

Figures reproduced from arXiv: 2501.11309 by the authors.

Figure 1
Figure 1. Illustration of Finer-CAM. Left: Sorted cosine similarity between linear classifier weights, averaged across all classes (details in the supplementary). Many pairs of classes are highly similar, yet neural networks can effectively distinguish them to achieve high fine-grained classification accuracy. Middle: Standard CAM methods highlight main regions contributing to the target class’s logit value, inadvertently inc… view at source ↗
Figure 2
Figure 2. Finer-CAM can be extended to multi-modal zero￾shot models to accurately highlight or mask out specific concepts. the target class’s logit value to its difference with a refer￾ence class’s logit—applicable to both gradient- and score￾based CAM methods. This comparison effectively uncov￾ers key distinctions between similar classes and highlights the corresponding regions in the images. Notably, Finer￾CAM does not requ… view at source ↗
Figure 3
Figure 3. The pipeline of the proposed Finer-CAM method, with Grad-CAM as the baseline. An image is first passed through the en￾coder blocks and the subsequent linear classifier to acquire feature maps at the desired network layer and the prediction logits, respectively. Different from standard Grad-CAM, we calculate the gradients of the logit difference between the target class and a visually similar class. In this way, the … view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: The visualization comparison between the proposed Finer-CAM and baseline CAM methods. For each group, we show the target image, one example image from the most similar class, baseline CAM, and Finer-CAM’s results. Finer-CAM localizes and emphasizes the discriminative d…
Figure 5
Figure 5. Figure 5: The saliency maps by Grad-CAM and Finer-CAM with deletion curves. In each group, the top-left is the target im￾age, while the bottom-left is an example image from the most sim￾ilar class. In addition to the prediction confidence of the target class, we also show the cu…
Figure 6
Figure 6. Figure 6: (a) Among images of the same species, Finer-CAM can generate saliency maps that consistently emphasize the same traits. [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Comparing the target class with different similar classes leads to a variety of activations. “None” denotes the baseline Grad-CAM without comparison. The class names are: Andean Siskin, Patagonian Sierra Finch, Canary, and Bananaquit. Consistency. Explainable methods a…
Figure 8
Figure 8. Figure 8: With a large comparison strength γ, the activation only focuses on fine-grained details. On the contrary, a small γ leads to coarse activation covering the entire object. on different cues for the target class. We first visualize the produced activation maps in [PITH_…
Figure 9
Figure 9. Figure 9: (a) Finer-CAM can be applied to verify if the classifier learns faithful knowledge of the trait differences between classes. For [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: Visualization of some failure cases where Finer-CAM cannot produce better saliency maps than the Grad-CAM baseline. capable of grounding the target object in the images. How￾ever, when asked to localize fine-grained traits or object parts, it still yields activations …
Figure 11
Figure 11. Figure 11: Visualization of multi-modal localization of fine-grained traits or object parts. For each original image, we aim to locate the target concept. By emphasizing the difference between the target concept and the original concept (“bird” or “car”), Finer-CAM accurately lo…
Figure 12
Figure 12. Figure 12: Qualitative comparison of the saliency maps generated by baseline CAM methods (Grad-CAM [37], Layer-CAM [17], and Score-CAM [41]), the proposed Finer-CAM applied on these three baselines, and other XAI methods (RISE [30] and Mask [12]). 4 [PITH_FULL_IMAGE:figures/ful…
Figure 13
Figure 13. Figure 13: Visualization of the extrapolation and reverse comparing cases with Grad-CAM as the baseline. The first two columns show the target image from class c and an image from the similar class d. The next two rows show the saliency maps generated by Grad-CAM and Finer-CAM. …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 40 canonical work pages

  1. [1]

    Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai

    Alejandro Barredo Arrieta, Natalia D ´ıaz-Rodr´ıguez, Javier Del Ser, Adrien Bennetot, Siham Tabik, Alberto Barbado, Salvador Garc´ıa, Sergio Gil-L ´opez, Daniel Molina, Richard Benjamins, et al. Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai. Information fusion, 58:82–115, 2020. 2

    work page 2020

  2. [2]

    On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation

    Sebastian Bach, Alexander Binder, Gr ´egoire Montavon, Frederick Klauschen, Klaus-Robert M ¨uller, and Wojciech Samek. On pixel-wise explanations for non-linear classifier decisions by layer-wise relevance propagation. PloS one, 10 (7):e0130140, 2015. 5

    work page 2015

  3. [3]

    How to explain individual classification decisions

    David Baehrens, Timon Schroeter, Stefan Harmeling, Mo- toaki Kawanabe, Katja Hansen, and Klaus-Robert M ¨uller. How to explain individual classification decisions. JMLR, 11:1803–1831, 2010. 3

    work page 2010

  4. [4]

    Grounding everything: Emerging localiza- tion properties in vision-language transformers

    Walid Bousselham, Felix Petersen, Vittorio Ferrari, and Hilde Kuehne. Grounding everything: Emerging localiza- tion properties in vision-language transformers. In Proceed- ings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3828–3837, 2024. 3

    work page 2024

  5. [5]

    Concise explanations of neural net- works using adversarial training

    Prasad Chalasani, Jiefeng Chen, Amrita Roy Chowdhury, Xi Wu, and Somesh Jha. Concise explanations of neural net- works using adversarial training. In ICML, pages 1383–

    work page

  6. [6]

    Grad-cam++: General- ized gradient-based visual explanations for deep convolu- tional networks

    Aditya Chattopadhay, Anirban Sarkar, Prantik Howlader, and Vineeth N Balasubramanian. Grad-cam++: General- ized gradient-based visual explanations for deep convolu- tional networks. In WACV, pages 839–847. IEEE, 2018. 3

    work page 2018

  7. [7]

    Transformer inter- pretability beyond attention visualization

    Hila Chefer, Shir Gur, and Lior Wolf. Transformer inter- pretability beyond attention visualization. In CVPR, pages 782–791, 2021. 2

    2021

  8. [8]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In CVPR, pages 248–255. Ieee, 2009. 1

    work page 2009

Show all 48 references
  1. [9]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition ...

    2021

  2. [10]

    Explainable and interpretable models in computer vision and machine learning

    Hugo Jair Escalante, Sergio Escalera, Isabelle Guyon, Xavier Bar´o, Ya˘gmur G¨uc ¸l¨ut¨urk, Umut G ¨uc ¸l¨u, Marcel van Gerven, and Rob van Lier. Explainable and interpretable models in computer vision and machine learning. Springer, 2018. 2

    2018

  3. [11]

    Un- derstanding deep networks via extremal perturbations and smooth masks

    Ruth Fong, Mandela Patrick, and Andrea Vedaldi. Un- derstanding deep networks via extremal perturbations and smooth masks. In ICCV, pages 2950–2958, 2019. 3

    2019

  4. [12]

    Interpretable explanations of black boxes by meaningful perturbation

    Ruth C Fong and Andrea Vedaldi. Interpretable explanations of black boxes by meaningful perturbation. In ICCV, pages 3429–3437, 2017. 2, 3, 4

    2017

  5. [13]

    Axiom-based grad-cam: Towards accurate visualization and explanation of cnns

    Ruigang Fu, Qingyong Hu, Xiaohu Dong, Yulan Guo, Yinghui Gao, and Biao Li. Axiom-based grad-cam: Towards accurate visualization and explanation of cnns. In BMVC, pages 1–13, 2020. 3

    2020

  6. [14]

    Deletion and insertion tests in regression models

    Naofumi Hama, Masayoshi Mase, and Art B Owen. Deletion and insertion tests in regression models. Journal of Machine Learning Research, 24(290):1–38, 2023. 5

    2023

  7. [15]

    Delving deep into rectifiers: Surpassing human-level perfor- mance on imagenet classification

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Delving deep into rectifiers: Surpassing human-level perfor- mance on imagenet classification. In ICCV, pages 1026– 1034, 2015. 1

    2015

  8. [16]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, pages 770–778, 2016. 1

    2016

  9. [17]

    Layercam: Exploring hierarchical class activation maps for localization

    Peng-Tao Jiang, Chang-Bin Zhang, Qibin Hou, Ming-Ming Cheng, and Yunchao Wei. Layercam: Exploring hierarchical class activation maps for localization. TIP, 30:5875–5888,

  10. [18]

    Adam: A method for stochastic opti- mization

    Diederik P Kingma. Adam: A method for stochastic opti- mization. arXiv preprint arXiv:1412.6980, 2014. 5

    2014 arXiv

  11. [19]

    Navigating the maze of explainable ai: A systematic approach to evaluat- ing methods and metrics

    Lukas Klein, Carsten T L ¨uth, Udo Schlegel, Till J Bungert, Mennatallah El-Assady, and Paul F J ¨ager. Navigating the maze of explainable ai: A systematic approach to evaluat- ing methods and metrics. arXiv preprint arXiv:2409.16756,

    arXiv

  12. [20]

    3d object representations for fine-grained categorization

    Jonathan Krause, Michael Stark, Jia Deng, and Li Fei-Fei. 3d object representations for fine-grained categorization. In ICCVW, pages 554–561, 2013. 3, 5, 6, 1, 2

    2013

  13. [21]

    Lundberg and Su-In Lee

    Scott M. Lundberg and Su-In Lee. A unified approach to in- terpreting model predictions. In NeurIPS, page 4768–4777,

  14. [22]

    A Unified Approach to Interpreting Model Predictions

    Scott M Lundberg and Su-In Lee. A Unified Approach to Interpreting Model Predictions. In NeurIPS, 2017. 3

    2017

  15. [23]

    Fine-grained visual classi- fication of aircraft

    Subhransu Maji, Esa Rahtu, Juho Kannala, Matthew Blaschko, and Andrea Vedaldi. Fine-grained visual classi- fication of aircraft. arXiv preprint arXiv:1306.5151 , 2013. 3, 5, 1

    2013 arXiv

  16. [24]

    Fish-vista: A multi-purpose dataset for understanding & identification of traits from images

    Kazi Sajeed Mehrab, M Maruf, Arka Daw, Harish Babu Manogaran, Abhilash Neog, Mridul Khurana, Bahadir Al- tintas, Yasin Bakis, Elizabeth G Campolongo, Matthew J Thompson, et al. Fish-vista: A multi-purpose dataset for understanding & identification of traits from images. arXiv ...

    2024 arXiv

  17. [25]

    Explainable artificial intelligence: a comprehensive review

    Dang Minh, H Xiang Wang, Y Fen Li, and Tan N Nguyen. Explainable artificial intelligence: a comprehensive review. Artificial Intelligence Review, pages 1–66, 2022. 2

    2022

  18. [26]

    Eigen-cam: Class activation map using principal compo- nents

    Mohammed Bany Muhammad and Mohammed Yeasin. Eigen-cam: Class activation map using principal compo- nents. In IJCNN, pages 1–7. IEEE, 2020. 3

    2020

  19. [27]

    Evet: enhancing visual explanations of deep neural networks using image transformations

    Youngrock Oh, Hyungsik Jung, Jeonghyung Park, and Min Soo Kim. Evet: enhancing visual explanations of deep neural networks using image transformations. In WACV, pages 3579–3587, 2021. 2

    2021

  20. [28]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193, 2023. 2, 3

    2023 arXiv

  21. [29]

    A simple interpretable transformer for fine-grained im- age classification and analysis

    Dipanjyoti Paul, Arpita Chowdhury, Xinqi Xiong, Feng-Ju Chang, David Edward Carlyn, Samuel Stevens, Kaiya L Provost, Anuj Karpatne, Bryan Carstens, Daniel Rubenstein, et al. A simple interpretable transformer for fine-grained im- age classification and analysis. In ICLR, 2024. 2 9

    2024

  22. [30]

    Rise: Randomized input sampling for explanation of black-box models

    V Petsiuk, A Das, and K Saenko. Rise: Randomized input sampling for explanation of black-box models. In BMVC, pages 1–13, 2018. 3, 5, 6, 4

    2018

  23. [31]

    Birds 525 species - image classification

    Gerald Piosenka. Birds 525 species - image classification

  24. [32]

    Learn- ing transferable visual models from natural language super- vision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learn- ing transferable visual models from natural language super- vision. In ICML, pages 8748–8763. PMLR, 2021. 2, 4, 5, 1

    2021

  25. [33]

    Ablation-cam: Visual explanations for deep convolutional network via gradient- free localization

    Harish Guruprasad Ramaswamy et al. Ablation-cam: Visual explanations for deep convolutional network via gradient- free localization. In WACV, pages 983–991, 2020. 3

    2020

  26. [34]

    There and back again: Revisiting backpropagation saliency methods

    Sylvestre-Alvise Rebuffi, Ruth Fong, Xu Ji, and Andrea Vedaldi. There and back again: Revisiting backpropagation saliency methods. In CVPR, pages 8839–8848, 2020. 3

    2020

  27. [35]

    ” why should i trust you?” explaining the predictions of any classifier

    Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. ” why should i trust you?” explaining the predictions of any classifier. In KDD, pages 1135–1144, 2016. 2, 3

    2016

  28. [36]

    Imagenet-21k pretraining for the masses,

    Tal Ridnik, Emanuel Ben-Baruch, Asaf Noy, and Lihi Zelnik-Manor. Imagenet-21k pretraining for the masses,

  29. [37]

    Grad-cam: Visual explanations from deep networks via gradient-based localization

    Ramprasaath R Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-cam: Visual explanations from deep networks via gradient-based localization. In ICCV, pages 618–626, 2017. 2, 3, 5, 6, 4

    2017

  30. [38]

    Smoothgrad: removing noise by adding noise

    Daniel Smilkov, Nikhil Thorat, Been Kim, Fernanda Vi ´egas, and Martin Wattenberg. Smoothgrad: removing noise by adding noise. arXiv preprint arXiv:1706.03825, 2017. 3

    2017 arXiv

  31. [39]

    Going deeper with convolutions

    Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. In CVPR, pages 1–9, 2015. 1

    2015

  32. [40]

    The caltech-ucsd birds-200-2011 dataset

    Catherine Wah, Steve Branson, Peter Welinder, Pietro Per- ona, and Serge Belongie. The caltech-ucsd birds-200-2011 dataset. 2011. 3, 5, 6, 8, 1

    2011

  33. [41]

    Score-cam: Score-weighted visual explanations for convolutional neural networks

    Haofan Wang, Zifan Wang, Mengnan Du, Fan Yang, Zijian Zhang, Sirui Ding, Piotr Mardziel, and Xia Hu. Score-cam: Score-weighted visual explanations for convolutional neural networks. In CVPRW, pages 24–25, 2020. 2, 3, 5, 6, 4

    2020

  34. [42]

    Fine-grained image analysis with deep learning: A survey

    Xiu-Shen Wei, Yi-Zhe Song, Oisin Mac Aodha, Jianxin Wu, Yuxin Peng, Jinhui Tang, Jian Yang, and Serge Belongie. Fine-grained image analysis with deep learning: A survey. TPAMI, 44(12):8927–8948, 2021. 3

    2021

  35. [43]

    Protopformer: Concentrating on prototypical parts in vision transform- ers for interpretable image recognition

    Mengqi Xue, Qihan Huang, Haofei Zhang, Lechao Cheng, Jie Song, Minghui Wu, and Mingli Song. Protopformer: Concentrating on prototypical parts in vision transform- ers for interpretable image recognition. arXiv preprint arXiv:2208.10431, 2022. 2

    2022

  36. [44]

    On the (in) fidelity and sensitivity of explanations

    Chih-Kuan Yeh, Cheng-Yu Hsieh, Arun Sai Suggala, David I Inouye, and Pradeep Ravikumar. On the (in) fidelity and sensitivity of explanations. In NeurIPS, pages 10967–10978,

  37. [45]

    A survey on deep learning-based fine-grained object classifi- cation and semantic segmentation

    Bo Zhao, Jiashi Feng, Xiao Wu, and Shuicheng Yan. A survey on deep learning-based fine-grained object classifi- cation and semantic segmentation. International Journal of Automation and Computing, 14(2):119–135, 2017. 3

    2017

  38. [46]

    Learn- ing multi-attention convolutional neural network for fine- grained image recognition

    Heliang Zheng, Jianlong Fu, Tao Mei, and Jiebo Luo. Learn- ing multi-attention convolutional neural network for fine- grained image recognition. In ICCV, pages 5209–5217,

  39. [47]

    Looking for the devil in the details: Learning trilinear atten- tion sampling network for fine-grained image recognition

    Heliang Zheng, Jianlong Fu, Zheng-Jun Zha, and Jiebo Luo. Looking for the devil in the details: Learning trilinear atten- tion sampling network for fine-grained image recognition. In CVPR, pages 5012–5021, 2019. 3

    2019

  40. [48]

    bird” as comparing pairs in the zero-shot classification setting. It shows the “ground truth

    Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba. Learning deep features for discrim- inative localization. In CVPR, pages 2921–2929, 2016. 2, 3 10 Finer-CAM : Spotting the Difference Reveals Finer Details for Visual Explanation Supplementary Materi...

    2016

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.