Ian Stavness

Masoomeh Aslahishahri, Jordan Ubbens, Ian Stavness

We present DARTS, a transformer model for reference-based image super-resolution. DARTS learns joint representations of two image distributions to enhance the content of low-resolution input images through matching correspondences learned from high-resolution reference images. Current state-of-the-art techniques in reference-based image super-resolution are based on a multi-network, multi-stage architecture. In this work, we adapt the double attention block from the GAN literature, processing the two visual streams separately and combining self-attention and cross-attention blocks through a gating attention strategy. Our work demonstrates how the attention mechanism can be adapted for the particular requirements of reference-based image super-resolution, significantly simplifying the architecture and training pipeline. We show that our transformer-based model performs competitively with state-of-the-art models, while maintaining a simpler overall architecture and training process. In particular, we obtain state-of-the-art on the SUN80 dataset, with a PSNR/SSIM of 29.83 / .809. These results show that attention alone is sufficient for the RSR task, without multiple purpose-built subnetworks, knowledge distillation, or multi-stage training.

Masoomeh Aslahishahri, Jordan Ubbens, Ian Stavness

In this paper, we propose HiTSR, a hierarchical transformer model for reference-based image super-resolution, which enhances low-resolution input images by learning matching correspondences from high-resolution reference images. Diverging from existing multi-network, multi-stage approaches, we streamline the architecture and training pipeline by incorporating the double attention block from GAN literature. Processing two visual streams independently, we fuse self-attention and cross-attention blocks through a gating attention strategy. The model integrates a squeeze-and-excitation module to capture global context from the input images, facilitating long-range spatial interactions within window-based attention blocks. Long skip connections between shallow and deep layers further enhance information flow. Our model demonstrates superior performance across three datasets including SUN80, Urban100, and Manga109. Specifically, on the SUN80 dataset, our model achieves PSNR/SSIM values of 30.24/0.821. These results underscore the effectiveness of attention mechanisms in reference-based image super-resolution. The transformer-based model attains state-of-the-art results without the need for purpose-built subnetworks, knowledge distillation, or multi-stage training, emphasizing the potency of attention in meeting reference-based image super-resolution requirements.

Tzu Ling Liu, Ian Stavness, Mrigank Rochan

Video Unsupervised Domain Adaptation (VUDA) poses a significant challenge in action recognition, requiring the adaptation of a model from a labeled source domain to an unlabeled target domain. Despite recent advances, existing VUDA methods often fall short of fully supervised performance, a key reason being the prevalence of static and uninformative backgrounds that exacerbate domain shifts. Additionally, prior approaches largely overlook computational efficiency, limiting real-world adoption. To address these issues, we propose Learnable Motion-Focused Tokenization (LMFT) for VUDA. LMFT tokenizes video frames into patch tokens and learns to discard low-motion, redundant tokens, primarily corresponding to background regions, while retaining motion-rich, action-relevant tokens for adaptation. Extensive experiments on three standard VUDA benchmarks across 21 domain adaptation settings show that our VUDA framework with LMFT achieves state-of-the-art performance while significantly reducing computational overhead. LMFT thus enables VUDA that is both effective and computationally efficient.

Shiori Sagawa, Pang Wei Koh, Tony Lee et al.

Machine learning systems deployed in the wild are often trained on a source distribution but deployed on a different target distribution. Unlabeled data can be a powerful point of leverage for mitigating these distribution shifts, as it is frequently much more available than labeled data and can often be obtained from distributions beyond the source distribution as well. However, existing distribution shift benchmarks with unlabeled data do not reflect the breadth of scenarios that arise in real-world applications. In this work, we present the WILDS 2.0 update, which extends 8 of the 10 datasets in the WILDS benchmark of distribution shifts to include curated unlabeled data that would be realistically obtainable in deployment. These datasets span a wide range of applications (from histology to wildlife conservation), tasks (classification, regression, and detection), and modalities (photos, satellite images, microscope slides, text, molecular graphs). The update maintains consistency with the original WILDS benchmark by using identical labeled training, validation, and test sets, as well as the evaluation metrics. On these datasets, we systematically benchmark state-of-the-art methods that leverage unlabeled data, including domain-invariant, self-training, and self-supervised methods, and show that their success on WILDS is limited. To facilitate method development and evaluation, we provide an open-source package that automates data loading and contains all of the model architectures and methods used in this paper. Code and leaderboards are available at https://wilds.stanford.edu.

Pang Wei Koh, Shiori Sagawa, Henrik Marklund et al.

Distribution shifts -- where the training distribution differs from the test distribution -- can substantially degrade the accuracy of machine learning (ML) systems deployed in the wild. Despite their ubiquity in the real-world deployments, these distribution shifts are under-represented in the datasets widely used in the ML community today. To address this gap, we present WILDS, a curated benchmark of 10 datasets reflecting a diverse range of distribution shifts that naturally arise in real-world applications, such as shifts across hospitals for tumor identification; across camera traps for wildlife monitoring; and across time and location in satellite imaging and poverty mapping. On each dataset, we show that standard training yields substantially lower out-of-distribution than in-distribution performance. This gap remains even with models trained by existing methods for tackling distribution shifts, underscoring the need for new methods for training models that are more robust to the types of distribution shifts that arise in practice. To facilitate method development, we provide an open-source package that automates dataset loading, contains default model architectures and hyperparameters, and standardizes evaluations. Code and leaderboards are available at https://wilds.stanford.edu.

Sara Mardanisamani, Farhad Maleki, Sara Hosseinzadeh Kassani et al.

Lodging, the permanent bending over of food crops, leads to poor plant growth and development. Consequently, lodging results in reduced crop quality, lowers crop yield, and makes harvesting difficult. Plant breeders routinely evaluate several thousand breeding lines, and therefore, automatic lodging detection and prediction is of great value aid in selection. In this paper, we propose a deep convolutional neural network (DCNN) architecture for lodging classification using five spectral channel orthomosaic images from canola and wheat breeding trials. Also, using transfer learning, we trained 10 lodging detection models using well-established deep convolutional neural network architectures. Our proposed model outperforms the state-of-the-art lodging detection methods in the literature that use only handcrafted features. In comparison to 10 DCNN lodging detection models, our proposed model achieves comparable results while having a substantially lower number of parameters. This makes the proposed model suitable for applications such as real-time classification using inexpensive hardware for high-throughput phenotyping pipelines. The GitHub repository at https://github.com/FarhadMaleki/LodgedNet contains code and models.

Najeeb Khan, Ian Stavness

Dropout is commonly used to help reduce overfitting in deep neural networks. Sparsity is a potentially important property of neural networks, but is not explicitly controlled by Dropout-based regularization. In this work, we propose Sparseout a simple and efficient variant of Dropout that can be used to control the sparsity of the activations in a neural network. We theoretically prove that Sparseout is equivalent to an $L_q$ penalty on the features of a generalized linear model and that Dropout is a special case of Sparseout for neural networks. We empirically demonstrate that Sparseout is computationally inexpensive and is able to control the desired level of sparsity in the activations. We evaluated Sparseout on image classification and language modelling tasks to see the effect of sparsity on these tasks. We found that sparsity of the activations is favorable for language modelling performance while image classification benefits from denser activations. Sparseout provides a way to investigate sparsity in state-of-the-art deep learning models. Source code for Sparseout could be found at \url{https://github.com/najeebkhan/sparseout}.

Yuemin Wang, Ian Stavness

This paper proposes a novel self-supervised learning method for semantic segmentation using selective masking image reconstruction as the pretraining task. Our proposed method replaces the random masking augmentation used in most masked image modelling pretraining methods. The proposed selective masking method selectively masks image patches with the highest reconstruction loss by breaking the image reconstruction pretraining into iterative steps to leverage the trained model's knowledge. We show on two general datasets (Pascal VOC and Cityscapes) and two weed segmentation datasets (Nassar 2020 and Sugarbeets 2016) that our proposed selective masking method outperforms the traditional random masking method and supervised ImageNet pretraining on downstream segmentation accuracy by 2.9% for general datasets and 2.5% for weed segmentation datasets. Furthermore, we found that our selective masking method significantly improves accuracy for the lowest-performing classes. Lastly, we show that using the same pretraining and downstream dataset yields the best result for low-budget self-supervised pretraining. Our proposed Selective Masking Image Reconstruction method provides an effective and practical solution to improve end-to-end semantic segmentation workflows, especially for scenarios that require limited model capacity to meet inference speed and computational resource requirements.

Keyhan Najafian, Farhad Maleki, Lingling Jin et al.

Instance segmentation is essential for applications such as automated monitoring of plant health, growth, and yield. However, extensive effort is required to create large-scale datasets with pixel-level annotations of each object instance for developing instance segmentation models that restrict the use of deep learning in these areas. This challenge is more significant in images with densely packed, self-occluded objects, which are common in agriculture. To address this challenge, we propose a semi-self-supervised learning approach that requires minimal manual annotation to develop a high-performing instance segmentation model. We design GLMask, an image-mask representation for the model to focus on shape, texture, and pattern while minimizing its dependence on color features. We develop a pipeline to generate semantic segmentation and then transform it into instance-level segmentation. The proposed approach substantially outperforms the conventional instance segmentation models, establishing a state-of-the-art wheat head instance segmentation model with mAP@50 of 98.5%. Additionally, we assessed the proposed methodology on the general-purpose Microsoft COCO dataset, achieving a significant performance improvement of over 12.6% mAP@50. This highlights that the utility of our proposed approach extends beyond precision agriculture and applies to other domains, specifically those with similar data characteristics.

Keyhan Najafian, Farhad Maleki, Lingling Jin et al.

Video object segmentation (VOS) -- predicting pixel-level regions for objects within each frame of a video -- is particularly challenging in agricultural scenarios, where videos of crops include hundreds of small, dense, and occluded objects (stems, leaves, flowers, pods) that sway and move unpredictably in the wind. Supervised training is the state-of-the-art for VOS, but it requires large, pixel-accurate, human-annotated videos, which are costly to produce for videos with many densely packed objects in each frame. To address these challenges, we proposed a semi-self-supervised spatiotemporal approach for dense-VOS (DVOS) using a diffusion-based method through multi-task (reconstruction and segmentation) learning. We train the model first with synthetic data that mimics the camera and object motion of real videos and then with pseudo-labeled videos. We evaluate our DVOS method for wheat head segmentation from a diverse set of videos (handheld, drone-captured, different field locations, and different growth stages -- spanning from Boot-stage to Wheat-mature and Harvest-ready). Despite using only a few manually annotated video frames, the proposed approach yielded a high-performing model, achieving a Dice score of 0.79 when tested on a drone-captured external test set. While our method was evaluated on wheat head segmentation, it can be extended to other crops and domains, such as crowd analysis or microscopic image analysis.

Etienne David, Mario Serouart, Daniel Smith et al.

The Global Wheat Head Detection (GWHD) dataset was created in 2020 and has assembled 193,634 labelled wheat heads from 4,700 RGB images acquired from various acquisition platforms and 7 countries/institutions. With an associated competition hosted in Kaggle, GWHD has successfully attracted attention from both the computer vision and agricultural science communities. From this first experience in 2020, a few avenues for improvements have been identified, especially from the perspective of data size, head diversity and label reliability. To address these issues, the 2020 dataset has been reexamined, relabeled, and augmented by adding 1,722 images from 5 additional countries, allowing for 81,553 additional wheat heads to be added. We now release a new version of the Global Wheat Head Detection (GWHD) dataset in 2021, which is bigger, more diverse, and less noisy than the 2020 version. The GWHD 2021 is now publicly available at http://www.global-wheat.com/ and a new data challenge has been organized on AIcrowd to make use of this updated dataset.

Etienne David, Franklin Ogidi, Wei Guo et al.

Data competitions have become a popular approach to crowdsource new data analysis methods for general and specialized data science problems. In plant phenotyping, data competitions have a rich history, and new outdoor field datasets have potential for new data competitions. We developed the Global Wheat Challenge as a generalization competition to see if solutions for wheat head detection from field images would work in different regions around the world. In this paper, we analyze the winning challenge solutions in terms of their robustness and the relative importance of model and data augmentation design decisions. We found that the design of the competition influence the selection of winning solutions and provide recommendations for future competitions in an attempt to garner more robust winning solutions.

Najeeb Khan, Ian Stavness

State-of-the-art computer vision models are rapidly increasing in capacity, where the number of parameters far exceeds the number required to fit the training set. This results in better optimization and generalization performance. However, the huge size of contemporary models results in large inference costs and limits their use on resource-limited devices. In order to reduce inference costs, convolutional filters in trained neural networks could be pruned to reduce the run-time memory and computational requirements during inference. However, severe post-training pruning results in degraded performance if the training algorithm results in dense weight vectors. We propose the use of Batch Bridgeout, a sparsity inducing stochastic regularization scheme, to train neural networks so that they could be pruned efficiently with minimal degradation in performance. We evaluate the proposed method on common computer vision models VGGNet, ResNet, and Wide-ResNet on the CIFAR image classification task. For all the networks, experimental results show that Batch Bridgeout trained networks achieve higher accuracy across a wide range of pruning intensities compared to Dropout and weight decay regularization.

Tewodros Ayalew, Jordan Ubbens, Ian Stavness

Supervised learning is often used to count objects in images, but for counting small, densely located objects, the required image annotations are burdensome to collect. Counting plant organs for image-based plant phenotyping falls within this category. Object counting in plant images is further challenged by having plant image datasets with significant domain shift due to different experimental conditions, e.g. applying an annotated dataset of indoor plant images for use on outdoor images, or on a different plant species. In this paper, we propose a domain-adversarial learning approach for domain adaptation of density map estimation for the purposes of object counting. The approach does not assume perfectly aligned distributions between the source and target datasets, which makes it more broadly applicable within general object counting and plant organ counting tasks. Evaluation on two diverse object counting tasks (wheat spikelets, leaves) demonstrates consistent performance on the target datasets across different classes of domain shift: from indoor-to-outdoor images and from species-to-species adaptation.

Jordan Ubbens, Tewodros Ayalew, Steve Shirtliffe et al.

Counting plant organs such as heads or tassels from outdoor imagery is a popular benchmark computer vision task in plant phenotyping, which has been previously investigated in the literature using state-of-the-art supervised deep learning techniques. However, the annotation of organs in field images is time-consuming and prone to errors. In this paper, we propose a fully unsupervised technique for counting dense objects such as plant organs. We use a convolutional network-based unsupervised segmentation method followed by two post-hoc optimization steps. The proposed technique is shown to provide competitive counting performance on a range of organ counting tasks in sorghum (S. bicolor) and wheat (T. aestivum) with no dataset-dependent tuning or modifications.

Shubhra Aich, Ian Stavness, Yasuhiro Taniguchi et al.

In this paper, we explore the idea of weight sharing over multiple scales in convolutional networks. Inspired by traditional computer vision approaches, we share the weights of convolution kernels over different scales in the same layers of the network. Although multi-scale feature aggregation and sharing inside convolutional networks are common in practice, none of the previous works address the issue of convolutional weight sharing. We evaluate our weight sharing scheme on two heterogeneous image recognition datasets - ImageNet (object recognition) and Places365-Standard (scene classification). With approximately 25% fewer parameters, our shared-weight ResNet model provides similar performance compared to baseline ResNets. Shared-weight models are further validated via transfer learning experiments on four additional image recognition datasets - Caltech256 and Stanford 40 Actions (object-centric) and SUN397 and MIT Inddor67 (scene-centric). Experimental results demonstrate significant redundancy in the vanilla implementations of the deeper networks, and also indicate that a shift towards increasing the receptive field per parameter may improve future convolutional network architectures.

Shubhra Aich, Ian Stavness

In this paper, we explore the problem of training one-look regression models for counting objects in datasets comprising a small number of high-resolution, variable-shaped images. We illustrate that conventional global average pooling (GAP) based models are unreliable due to the patchwise cancellation of true overestimates and underestimates for patchwise inference. To overcome this limitation and reduce overfitting caused by the training on full-resolution images, we propose to employ global sum pooling (GSP) instead of GAP or fully connected (FC) layers at the backend of a convolutional network. Although computationally equivalent to GAP, we show through comprehensive experimentation that GSP allows convolutional networks to learn the counting task as a simple linear mapping problem generalized over the input shape and the number of objects present. This generalization capability allows GSP to avoid both patchwise cancellation and overfitting by training on small patches and inference on full-resolution images as a whole. We evaluate our approach on four different aerial image datasets - two car counting datasets (CARPK and COWC), one crowd counting dataset (ShanghaiTech; parts A and B) and one new challenging dataset for wheat spike counting. Our GSP models improve upon the state-of-the-art approaches on all four datasets with a simple architecture. Also, GSP architectures trained with smaller-sized image patches exhibit better localization property due to their focus on learning from smaller regions while training.

Shubhra Aich, William van der Kamp, Ian Stavness

In this paper, we propose an efficient architecture for semantic image segmentation using the depth-to-space (D2S) operation. Our D2S model is comprised of a standard CNN encoder followed by a depth-to-space reordering of the final convolutional feature maps. Our approach eliminates the decoder portion of traditional encoder-decoder segmentation models and reduces the amount of computation almost by half. As a participant of the DeepGlobe Road Extraction competition, we evaluate our models on the corresponding road segmentation dataset. Our highly efficient D2S models exhibit comparable performance to standard segmentation models with much lower computational cost.

Najeeb Khan, Jawad Shah, Ian Stavness

A major challenge in training deep neural networks is overfitting, i.e. inferior performance on unseen test examples compared to performance on training examples. To reduce overfitting, stochastic regularization methods have shown superior performance compared to deterministic weight penalties on a number of image recognition tasks. Stochastic methods such as Dropout and Shakeout, in expectation, are equivalent to imposing a ridge and elastic-net penalty on the model parameters, respectively. However, the choice of the norm of weight penalty is problem dependent and is not restricted to $\{L_1,L_2\}$. Therefore, in this paper we propose the Bridgeout stochastic regularization technique and prove that it is equivalent to an $L_q$ penalty on the weights, where the norm $q$ can be learned as a hyperparameter from data. Experimental results show that Bridgeout results in sparse model weights, improved gradients and superior classification performance compared to Dropout and Shakeout on synthetic and real datasets.

Shubhra Aich, Ian Stavness

In this paper, we propose a simple and effective way to improve one-look regression models for object counting from images. We use class activation map visualizations to illustrate the drawbacks of learning a pure one-look regression model for a counting task. Based on these insights, we enhance one-look regression counting models by regulating activation maps from the final convolution layer of the network with coarse ground-truth activation maps generated from simple dot annotations. We call this strategy heatmap regulation (HR). We show that this simple enhancement effectively suppresses false detections generated by the corresponding one-look baseline model and also improves the performance in terms of false negatives. Evaluations are performed on four different counting datasets --- two for car counting (CARPK, PUCPR+), one for crowd counting (WorldExpo) and another for biological cell counting (VGG-Cells). Adding HR to a simple VGG front-end improves performance on all these benchmarks compared to a simple one-look baseline model and results in state-of-the-art performance for car counting.

Shubhra Aich, Anique Josuttes, Ilya Ovsyannikov et al.

In this paper, we investigate estimating emergence and biomass traits from color images and elevation maps of wheat field plots. We employ a state-of-the-art deconvolutional network for segmentation and convolutional architectures, with residual and Inception-like layers, to estimate traits via high dimensional nonlinear regression. Evaluation was performed on two different species of wheat, grown in field plots for an experimental plant breeding study. Our framework achieves satisfactory performance with mean and standard deviation of absolute difference of 1.05 and 1.40 counts for emergence and 1.45 and 2.05 for biomass estimation. Our results for counting wheat plants from field images are better than the accuracy reported for the similar, but arguably less difficult, task of counting leaves from indoor images of rosette plants. Our results for biomass estimation, even with a very small dataset, improve upon all previously proposed approaches in the literature.

Shubhra Aich, Ian Stavness

In this paper, we investigate the problem of counting rosette leaves from an RGB image, an important task in plant phenotyping. We propose a data-driven approach for this task generalized over different plant species and imaging setups. To accomplish this task, we use state-of-the-art deep learning architectures: a deconvolutional network for initial segmentation and a convolutional network for leaf counting. Evaluation is performed on the leaf counting challenge dataset at CVPPP-2017. Despite the small number of training samples in this dataset, as compared to typical deep learning image sets, we obtain satisfactory performance on segmenting leaves from the background as a whole and counting the number of leaves using simple data augmentation strategies. Comparative analysis is provided against methods evaluated on the previous competition datasets. Our framework achieves mean and standard deviation of absolute count difference of 1.62 and 2.30 averaged over all five test datasets.

Najeeb Khan, Ian Stavness

The motor control problem involves determining the time-varying muscle activation trajectories required to accomplish a given movement. Muscle redundancy makes motor control a challenging task: there are many possible activation trajectories that accomplish the same movement. Despite this redundancy, most movements are accomplished in highly stereotypical ways. For example, point-to-point reaching movements are almost universally performed with very similar smooth trajectories. Optimization methods are commonly used to predict muscle forces for measured movements. However, these approaches require computationally expensive simulations and are sensitive to the chosen optimality criteria and regularization. In this work, we investigate deep autoencoders for the prediction of muscle activation trajectories for point-to-point reaching movements. We evaluate our DNN predictions with simulated reaches and two methods to generate the muscle activations: inverse dynamics (ID) and optimal control (OC) criteria. We also investigate optimal network parameters and training criteria to improve the accuracy of the predictions.