Prompt-CAM: A Simpler Interpretable Transformer for Fine-Grained Analysis

We present a simple usage of pre-trained Vision Transformers (ViTs) for fine-grained analysis, aiming to identify and localize the traits that distinguish visually similar categories, such as different bird species or dog breeds. Pre-trained ViTs such as DINO have shown remarkable capabilities to extract localized, informative features. However, using saliency maps like Grad-CAM can hardly point out the traits: they often locate the whole object by a blurred, coarse heatmap, not traits. We propose a novel approach Prompt Class Attention Map (Prompt-CAM) to the rescue. Prompt-CAM learns class-specific prompts to a pre-trained ViT and uses the corresponding outputs for classification. To classify an image correctly, the true-class prompt must attend to the unique image patches not seen in other classes' images, i.e., traits. As such, the true class's multi-head attention maps reveal traits and their locations. Implementation-wise, Prompt-CAM is almost a free lunch by simply modifying the prediction head of Visual Prompt Tuning (VPT). This makes Prompt-CAM fairly easy to train and apply, sharply contrasting other interpretable methods that design specific models and training processes. It is even simpler than the recently published INterpretable TRansformer (INTR), whose encoder-decoder architecture prevents it from leveraging pre-trained ViTs. Extensive empirical studies on a dozen datasets from various domains (e.g., birds, fishes, insects, fungi, flowers, food, and cars) validate Prompt-CAM superior interpretation capability.

Lessons Learned from a Unifying Empirical Study of Parameter-Efficient Transfer Learning (PETL) in Visual Recognition

Parameter-efficient transfer learning (PETL) has attracted significant attention lately, due to the increasing size of pre-trained models and the need to fine-tune (FT) them for superior downstream performance. This community-wide enthusiasm has sparked a plethora of new methods. Nevertheless, a systematic study to understand their performance and suitable application scenarios is lacking, leaving questions like when to apply PETL and which method to use largely unanswered. In this paper, we conduct a unifying empirical study of representative PETL methods in the context of Vision Transformers. We systematically tune their hyper-parameters to fairly compare their accuracy on downstream tasks. Our study not only offers a valuable user guide but also unveils several new insights. First, if tuned carefully, different PETL methods can obtain quite similar accuracy in the low-shot benchmark VTAB-1K. This includes simple methods like FT the bias terms that were reported inferior. Second, though with similar accuracy, we find that PETL methods make different mistakes and high-confidence predictions, likely due to their different inductive biases. Such an inconsistency (or complementariness) opens up the opportunity for ensemble methods, and we make preliminary attempts at this. Third, going beyond the commonly used low-shot tasks, we find that PETL is also useful in many-shot regimes -- it achieves comparable and sometimes better accuracy than full FT, using much fewer learnable parameters. Last but not least, we investigate PETL's ability to preserve a pre-trained model's robustness to distribution shifts (e.g., a CLIP backbone). Perhaps not surprisingly, PETL methods outperform full FT alone. However, with weight-space ensembles, the fully FT model can achieve a better balance between downstream and out-of-distribution performance, suggesting a future research direction for PETL.

Fine-Tuning is Fine, if Calibrated

Fine-tuning is arguably the most straightforward way to tailor a pre-trained model (e.g., a foundation model) to downstream applications, but it also comes with the risk of losing valuable knowledge the model had learned in pre-training. For example, fine-tuning a pre-trained classifier capable of recognizing a large number of classes to master a subset of classes at hand is shown to drastically degrade the model's accuracy in the other classes it had previously learned. As such, it is hard to further use the fine-tuned model when it encounters classes beyond the fine-tuning data. In this paper, we systematically dissect the issue, aiming to answer the fundamental question, ''What has been damaged in the fine-tuned model?'' To our surprise, we find that the fine-tuned model neither forgets the relationship among the other classes nor degrades the features to recognize these classes. Instead, the fine-tuned model often produces more discriminative features for these other classes, even if they were missing during fine-tuning! {What really hurts the accuracy is the discrepant logit scales between the fine-tuning classes and the other classes}, implying that a simple post-processing calibration would bring back the pre-trained model's capability and at the same time unveil the feature improvement over all classes. We conduct an extensive empirical study to demonstrate the robustness of our findings and provide preliminary explanations underlying them, suggesting new directions for future theoretical analysis. Our code is available at https://github.com/OSU-MLB/Fine-Tuning-Is-Fine-If-Calibrated.

CompBench: A Comparative Reasoning Benchmark for Multimodal LLMs

The ability to compare objects, scenes, or situations is crucial for effective decision-making and problem-solving in everyday life. For instance, comparing the freshness of apples enables better choices during grocery shopping, while comparing sofa designs helps optimize the aesthetics of our living space. Despite its significance, the comparative capability is largely unexplored in artificial general intelligence (AGI). In this paper, we introduce CompBench, a benchmark designed to evaluate the comparative reasoning capability of multimodal large language models (MLLMs). CompBench mines and pairs images through visually oriented questions covering eight dimensions of relative comparison: visual attribute, existence, state, emotion, temporality, spatiality, quantity, and quality. We curate a collection of around 40K image pairs using metadata from diverse vision datasets and CLIP similarity scores. These image pairs span a broad array of visual domains, including animals, fashion, sports, and both outdoor and indoor scenes. The questions are carefully crafted to discern relative characteristics between two images and are labeled by human annotators for accuracy and relevance. We use CompBench to evaluate recent MLLMs, including GPT-4V(ision), Gemini-Pro, and LLaVA-1.6. Our results reveal notable shortcomings in their comparative abilities. We believe CompBench not only sheds light on these limitations but also establishes a solid foundation for future enhancements in the comparative capability of MLLMs.

Holistic Transfer: Towards Non-Disruptive Fine-Tuning with Partial Target Data

We propose a learning problem involving adapting a pre-trained source model to the target domain for classifying all classes that appeared in the source data, using target data that covers only a partial label space. This problem is practical, as it is unrealistic for the target end-users to collect data for all classes prior to adaptation. However, it has received limited attention in the literature. To shed light on this issue, we construct benchmark datasets and conduct extensive experiments to uncover the inherent challenges. We found a dilemma -- on the one hand, adapting to the new target domain is important to claim better performance; on the other hand, we observe that preserving the classification accuracy of classes missing in the target adaptation data is highly challenging, let alone improving them. To tackle this, we identify two key directions: 1) disentangling domain gradients from classification gradients, and 2) preserving class relationships. We present several effective solutions that maintain the accuracy of the missing classes and enhance the overall performance, establishing solid baselines for holistic transfer of pre-trained models with partial target data.

Segment Anything Model (SAM) Enhanced Pseudo Labels for Weakly Supervised Semantic Segmentation

Weakly Supervised Semantic Segmentation (WSSS) with only image-level supervision has garnered increasing attention due to its low annotation cost compared to pixel-level annotation. Most existing methods rely on Class Activation Maps (CAM) to generate pixel-level pseudo labels for supervised training. However, it is well known that CAM often suffers from partial activation -- activating the most discriminative part instead of the entire object area, and false activation -- unnecessarily activating the background around the object. In this study, we introduce a simple yet effective approach to address these limitations by harnessing the recently released Segment Anything Model (SAM) to generate higher-quality pseudo labels with CAM. SAM is a segmentation foundation model that demonstrates strong zero-shot ability in partitioning images into segments but lacks semantic labels for these regions. To circumvent this, we employ pseudo labels for a specific class as the signal to select the most relevant masks and label them to generate the refined pseudo labels for this class. The segments generated by SAM are highly precise, leading to substantial improvements in partial and false activation. Moreover, existing post-processing modules for producing pseudo labels, such as AffinityNet, are often computationally heavy, with a significantly long training time. Surprisingly, we discovered that using the initial CAM with SAM can achieve on-par performance as the post-processed pseudo label generated from these modules with much less computational cost. Our approach is highly versatile and capable of seamless integration into existing WSSS models without modification to base networks or pipelines. Despite its simplicity, our approach improves the mean Intersection over Union (mIoU) of pseudo labels from five state-of-the-art WSSS methods by 6.2\% on average on the PASCAL VOC 2012 dataset.

Visual Query Tuning: Towards Effective Usage of Intermediate Representations for Parameter and Memory Efficient Transfer Learning

Intermediate features of a pre-trained model have been shown informative for making accurate predictions on downstream tasks, even if the model backbone is kept frozen. The key challenge is how to utilize these intermediate features given their gigantic amount. We propose visual query tuning (VQT), a simple yet effective approach to aggregate intermediate features of Vision Transformers. Through introducing a handful of learnable ``query'' tokens to each layer, VQT leverages the inner workings of Transformers to ``summarize'' rich intermediate features of each layer, which can then be used to train the prediction heads of downstream tasks. As VQT keeps the intermediate features intact and only learns to combine them, it enjoys memory efficiency in training, compared to many other parameter-efficient fine-tuning approaches that learn to adapt features and need back-propagation through the entire backbone. This also suggests the complementary role between VQT and those approaches in transfer learning. Empirically, VQT consistently surpasses the state-of-the-art approach that utilizes intermediate features for transfer learning and outperforms full fine-tuning in many cases. Compared to parameter-efficient approaches that adapt features, VQT achieves much higher accuracy under memory constraints. Most importantly, VQT is compatible with these approaches to attain even higher accuracy, making it a simple add-on to further boost transfer learning.

TransCAM: Transformer Attention-based CAM Refinement for Weakly Supervised Semantic Segmentation

Weakly supervised semantic segmentation (WSSS) with only image-level supervision is a challenging task. Most existing methods exploit Class Activation Maps (CAM) to generate pixel-level pseudo labels for supervised training. However, due to the local receptive field of Convolution Neural Networks (CNN), CAM applied to CNNs often suffers from partial activation -- highlighting the most discriminative part instead of the entire object area. In order to capture both local features and global representations, the Conformer has been proposed to combine a visual transformer branch with a CNN branch. In this paper, we propose TransCAM, a Conformer-based solution to WSSS that explicitly leverages the attention weights from the transformer branch of the Conformer to refine the CAM generated from the CNN branch. TransCAM is motivated by our observation that attention weights from shallow transformer blocks are able to capture low-level spatial feature similarities while attention weights from deep transformer blocks capture high-level semantic context. Despite its simplicity, TransCAM achieves a new state-of-the-art performance of 69.3% and 69.6% on the respective PASCAL VOC 2012 validation and test sets, showing the effectiveness of transformer attention-based refinement of CAM for WSSS.

Unintended Bias in Language Model-driven Conversational Recommendation

Conversational Recommendation Systems (CRSs) have recently started to leverage pretrained language models (LM) such as BERT for their ability to semantically interpret a wide range of preference statement variations. However, pretrained LMs are well-known to be prone to intrinsic biases in their training data, which may be exacerbated by biases embedded in domain-specific language data(e.g., user reviews) used to fine-tune LMs for CRSs. We study a recently introduced LM-driven recommendation backbone (termed LMRec) of a CRS to investigate how unintended bias i.e., language variations such as name references or indirect indicators of sexual orientation or location that should not affect recommendations manifests in significantly shifted price and category distributions of restaurant recommendations. The alarming results we observe strongly indicate that LMRec has learned to reinforce harmful stereotypes through its recommendations. For example, offhand mention of names associated with the black community significantly lowers the price distribution of recommended restaurants, while offhand mentions of common male-associated names lead to an increase in recommended alcohol-serving establishments. These and many related results presented in this work raise a red flag that advances in the language handling capability of LM-drivenCRSs do not come without significant challenges related to mitigating unintended bias in future deployed CRS assistants with a potential reach of hundreds of millions of end-users.

Supervised Contrastive Replay: Revisiting the Nearest Class Mean Classifier in Online Class-Incremental Continual Learning

Online class-incremental continual learning (CL) studies the problem of learning new classes continually from an online non-stationary data stream, intending to adapt to new data while mitigating catastrophic forgetting. While memory replay has shown promising results, the recency bias in online learning caused by the commonly used Softmax classifier remains an unsolved challenge. Although the Nearest-Class-Mean (NCM) classifier is significantly undervalued in the CL community, we demonstrate that it is a simple yet effective substitute for the Softmax classifier. It addresses the recency bias and avoids structural changes in the fully-connected layer for new classes. Moreover, we observe considerable and consistent performance gains when replacing the Softmax classifier with the NCM classifier for several state-of-the-art replay methods. To leverage the NCM classifier more effectively, data embeddings belonging to the same class should be clustered and well-separated from those with a different class label. To this end, we contribute Supervised Contrastive Replay (SCR), which explicitly encourages samples from the same class to cluster tightly in embedding space while pushing those of different classes further apart during replay-based training. Overall, we observe that our proposed SCR substantially reduces catastrophic forgetting and outperforms state-of-the-art CL methods by a significant margin on a variety of datasets.