Recent advances in large language and vision-language models have enabled zero-shot inference, allowing models to solve new tasks without task-specific training. Various adaptation techniques such as prompt engineering, In-Context Learning (ICL), and supervised fine-tuning can further enhance the model’s performance on a downstream task, but they require substantial manual effort to construct effective prompts or labeled examples. In this work, we introduce a joint inference framework for fully unsupervised adaptation, eliminating the need for manual prompt engineering and labeled examples. Unlike zero-shot inference, which makes independent predictions, the joint inference makes predictions simultaneously for all inputs in a given task. Since direct joint inference involves computationally expensive optimization, we develop efficient approximation techniques, leading to two unsupervised adaptation methods: unsupervised fine-tuning and unsupervised ICL. We demonstrate the effectiveness of our methods across diverse tasks and models, including language-only Llama-3.1 on natural language processing tasks, reasoning-oriented Qwen2.5-Math on grade school math problems, vision-language OpenFlamingo on vision tasks, and the API-only access GPT-4o model on massive multi-discipline tasks. Our experiments demonstrate substantial improvements over the standard zero-shot approach, including 39% absolute improvement on the challenging GSM8K math reasoning dataset. Remarkably, despite being fully unsupervised, our framework often performs on par with supervised approaches that rely on ground truth labels.

We introduce FlexTok, a tokenizer that projects 2D images into variable-length, ordered 1D token sequences. For example, a 256x256 image can be resampled into anywhere from 1 to 256 discrete tokens, hierarchically and semantically compressing its information. By training a rectified flow model as the decoder and using nested dropout, FlexTok produces plausible reconstructions regardless of the chosen token sequence length. We evaluate our approach in an autoregressive generation setting using a simple GPT-style Transformer. On ImageNet, this approach achieves an FID$<$2 across 8 to 128 tokens, outperforming TiTok and matching state-of-the-art methods with far fewer tokens. We further extend the model to support to text-conditioned image generation and examine how FlexTok relates to traditional 2D tokenization. A key finding is that FlexTok enables next-token prediction to describe images in a coarse-to-fine "visual vocabulary", and that the number of tokens to generate depends on the complexity of the generation task.

Supervised learning datasets may contain multiple cues that explain the training set equally well, i.e., learning any of them would lead to the correct predictions on the training data. However, many of them can be spurious, i.e., lose their predictive power under a distribution shift and consequently fail to generalize to out-of-distribution (OOD) data. Recently developed "diversification" methods (Lee et al., 2023; Pagliardini et al., 2023) approach this problem by finding multiple diverse hypotheses that rely on different features. This paper aims to study this class of methods and identify the key components contributing to their OOD generalization abilities. We show that (1) diversification methods are highly sensitive to the distribution of the unlabeled data used for diversification and can underperform significantly when away from a method-specific sweet spot. (2) Diversification alone is insufficient for OOD generalization. The choice of the used learning algorithm, e.g., the model's architecture and pretraining, is crucial. In standard experiments (classification on Waterbirds and Office-Home datasets), using the second-best choice leads to an up to 20\% absolute drop in accuracy. (3) The optimal choice of learning algorithm depends on the unlabeled data and vice versa i.e. they are co-dependent. (4) Finally, we show that, in practice, the above pitfalls cannot be alleviated by increasing the number of diverse hypotheses, the major feature of diversification methods. These findings provide a clearer understanding of the critical design factors influencing the OOD generalization abilities of diversification methods. They can guide practitioners in how to use the existing methods best and guide researchers in developing new, better ones.

Current multimodal and multitask foundation models, like 4M or UnifiedIO, show promising results. However, their out-of-the-box abilities to accept diverse inputs and perform diverse tasks are limited by the (usually small) number of modalities and tasks they are trained on. In this paper, we develop a single any-to-any model trained on tens of highly diverse modalities and by performing co-training on large-scale multimodal datasets and text corpora. This includes training on images and text along with several semantic and geometric modalities, feature maps from recent state of the art models like DINOv2 and ImageBind, pseudo labels of specialist models like SAM and 4DHumans, and a range of new modalities that allow for novel ways to interact with the model and steer the generation, for example, image metadata or color palettes. A crucial step in this process is performing discrete tokenization on various modalities, whether they are image-like, neural network feature maps, vectors, structured data like instance segmentation or human poses, or data that can be represented as text. Through this, we show the possibility of training one model to solve at least 3x more tasks/modalities than existing models and doing so without a loss in performance. In addition, this enables more fine-grained and controllable multimodal generation capabilities and allows studying the distillation of models trained on diverse data and objectives into one unified model. We scale the training to a three billion parameter and different datasets. The multimodal models and training code are open sourced at https://4m.epfl.ch/.

Effectively localizing an agent in a realistic, noisy setting is crucial for many embodied vision tasks. Visual Odometry (VO) is a practical substitute for unreliable GPS and compass sensors, especially in indoor environments. While SLAM-based methods show a solid performance without large data requirements, they are less flexible and robust w.r.t. to noise and changes in the sensor suite compared to learning-based approaches. Recent deep VO models, however, limit themselves to a fixed set of input modalities, e.g., RGB and depth, while training on millions of samples. When sensors fail, sensor suites change, or modalities are intentionally looped out due to available resources, e.g., power consumption, the models fail catastrophically. Furthermore, training these models from scratch is even more expensive without simulator access or suitable existing models that can be fine-tuned. While such scenarios get mostly ignored in simulation, they commonly hinder a model's reusability in real-world applications. We propose a Transformer-based modality-invariant VO approach that can deal with diverse or changing sensor suites of navigation agents. Our model outperforms previous methods while training on only a fraction of the data. We hope this method opens the door to a broader range of real-world applications that can benefit from flexible and learned VO models.

We propose a method for adapting neural networks to distribution shifts at test-time. In contrast to training-time robustness mechanisms that attempt to anticipate and counter the shift, we create a closed-loop system and make use of test-time feedback signal to adapt a network on the fly. We show that this loop can be effectively implemented using a learning-based function, which realizes an amortized optimizer for the network. This leads to an adaptation method, named Rapid Network Adaptation (RNA), that is notably more flexible and orders of magnitude faster than the baselines. Through a broad set of experiments using various adaptation signals and target tasks, we study the generality, efficiency, and flexibility of this method. We perform the evaluations using various datasets (Taskonomy, Replica, ScanNet, Hypersim, COCO, ImageNet), tasks (depth, optical flow, semantic segmentation, classification), and distribution shifts (Cross-datasets, 2D and 3D Common Corruptions) with promising results.

Current machine learning models for vision are often highly specialized and limited to a single modality and task. In contrast, recent large language models exhibit a wide range of capabilities, hinting at a possibility for similarly versatile models in computer vision. In this paper, we take a step in this direction and propose a multimodal training scheme called 4M. It consists of training a single unified Transformer encoder-decoder using a masked modeling objective across a wide range of input/output modalities – including text, images, geometric, and semantic modalities, as well as neural network feature maps. 4M achieves scalability by unifying the representation space of all modalities through mapping them into discrete tokens and performing multimodal masked modeling on a small randomized subset of tokens. 4M leads to models that exhibit several key capabilities: (1) they can perform a diverse set of vision tasks out of the box, (2) they excel when fine-tuned for unseen downstream tasks or new input modalities, and (3) they can function as a generative model that can be conditioned on arbitrary modalities, enabling a wide variety of expressive multimodal editing capabilities with remarkable flexibility. Through experimental analyses, we demonstrate the potential of 4M for training versatile and scalable foundation models for vision tasks, setting the stage for further exploration in multimodal learning for vision and other domains.

We introduce a set of image transformations that can be used as corruptions to evaluate the robustness of models as well as data augmentation mechanisms for training neural networks. The primary distinction of the proposed transformations is that, unlike existing approaches such as Common Corruptions [27], the geometry of the scene is incorporated in the transformations - thus leading to corruptions that are more likely to occur in the real world. We also introduce a set of semantic corruptions (e.g. natural object occlusions. See Fig. 1). We show these transformations are ‘efficient’ (can be computed on-the-fly), ‘extendable’ (can be applied on most image datasets), expose vulnerability of existing models, and can effectively make models more robust when employed as ‘3D data augmentation’ mechanisms. The evaluations on several tasks and datasets suggest incorporating 3D information into benchmarking and training opens up a promising direction for robustness research.

When developing deep learning models, we usually decide what task we want to solve then search for a model that generalizes well on the task. An intriguing question would be: what if, instead of fixing the task and searching in the model space, we fix the model and search in the task space? Can we find tasks that the model generalizes on? How do they look, or do they indicate anything? These are the questions we address in this paper. We propose a task discovery framework that automatically finds examples of such tasks via optimizing a generalization-based quantity called agreement score. We demonstrate that one set of images can give rise to many tasks on which neural networks generalize well. These tasks are a reflection of the inductive biases of the learning framework and the statistical patterns present in the data, thus they can make a useful tool for analyzing the neural networks and their biases. As an example, we show that the discovered tasks can be used to automatically create ''adversarial train-test splits'' which make a model fail at test time, without changing the pixels or labels, but by only selecting how the datapoints should be split between the train and test sets. We end with a discussion on human-interpretability of the discovered tasks.

To achieve autonomy in a priori unknown real-world scenarios, agents should be able to: i) act from high-dimensional sensory observations (e.g., images), ii) learn from past experience to adapt and improve, and iii) be capable of long horizon planning. Classical planning algorithms (e.g. PRM, RRT) are proficient at handling long-horizon planning. Deep learning based methods in turn can provide the necessary representations to address the others, by modeling statistical contingencies between observations. In this direction, we introduce a general-purpose planning algorithm called PALMER that combines classical sampling-based planning algorithms with learning-based perceptual representations. For training these perceptual representations, we combine Q-learning with contrastive representation learning to create a latent space where the distance between the embeddings of two states captures how easily an optimal policy can traverse between them. For planning with these perceptual representations, we re-purpose classical sampling-based planning algorithms to retrieve previously observed trajectory segments from a replay buffer and restitch them into approximately optimal paths that connect any given pair of start and goal states. This creates a tight feedback loop between representation learning, memory, reinforcement learning, and sampling-based planning. The end result is an experiential framework for long-horizon planning that is significantly more robust and sample efficient compared to existing methods.

Abstraction is at the heart of sketching due to the simple and minimal nature of line drawings. Abstraction entails identifying the essential visual properties of an object or scene, which requires semantic understanding and prior knowledge of high-level concepts. Abstract depictions are therefore challenging for artists, and even more so for machines. We present CLIPasso, an object sketching method that can achieve different levels of abstraction, guided by geometric and semantic simplifications. While sketch generation methods often rely on explicit sketch datasets for training, we utilize the remarkable ability of CLIP (Contrastive-Language-Image-Pretraining) to distill semantic concepts from sketches and images alike. We define a sketch as a set of Bézier curves and use a differentiable rasterizer to optimize the parameters of the curves directly with respect to a CLIP-based perceptual loss. The abstraction degree is controlled by varying the number of strokes. The generated sketches demonstrate multiple levels of abstraction while maintaining recognizability, underlying structure, and essential visual components of the subject drawn.

This paper introduces a pipeline to parametrically sample and render static multi-task vision datasets from comprehensive 3D scans from the real-world. In addition to enabling interesting lines of research, we show the tooling and generated data suffice to train robust vision models. Familiar architectures trained on a generated starter dataset reached state-of-the-art performance on multiple common vision tasks and benchmarks, despite having seen no benchmark or non-pipeline data. The depth estimation network outperforms MiDaS and the surface normal estimation network is the first to achieve human-level performance for in-the-wild surface normal estimation—at least according to one metric on the OASIS benchmark. The Dockerized pipeline with CLI, the (mostly python) code, PyTorch dataloaders for the generated data, the generated starter dataset, download scripts and other utilities are all available ${\color{Magenta}through}\;{\color{Magenta}our}\;{\color{Magenta}project}\;{\color{Magenta}website}$.

We present a method for making neural network predictions robust to shifts from the training data distribution. The proposed method is based on making predictions via a diverse set of cues (called ‘middle domains’) and ensembling them into one strong prediction. The premise of the idea is that predictions made via different cues respond differently to a distribution shift, hence one should be able to merge them into one robust final prediction. We perform the merging in a straightforward but principled manner based on the uncertainty associated with each prediction. The evaluations are performed using multiple tasks and datasets (Taskonomy, Replica, ImageNet, CIFAR) under a wide range of adversarial and non-adversarial distribution shifts which demonstrate the proposed method is considerably more robust than its standard learning counterpart, conventional deep ensembles, and several other baselines.

Visual perception entails solving a wide set of tasks (e.g., object detection, depth estimation, etc). The predictions made for different tasks out of one image are not independent, and therefore, are expected to be 'consistent'. We propose a flexible and fully computational framework for learning while enforcing Cross-Task Consistency (X-TAC). The proposed formulation is based on 'inference path invariance' over an arbitrary graph of prediction domains. We observe that learning with cross-task consistency leads to more accurate predictions, better generalization to out-of-distribution samples, and improved sample efficiency. This framework also leads to a powerful unsupervised quantity, called 'Consistency Energy, based on measuring the intrinsic consistency of the system. Consistency Energy well correlates with the supervised error (r=0.67), thus it can be employed as an unsupervised robustness metric as well as for detection of out-of-distribution inputs (AUC=0.99). The evaluations were performed on multiple datasets, including Taskonomy, Replica, CocoDoom, and ApolloScape.

Many computer vision applications require solving multiple tasks in real-time. A neural network can be trained to solve multiple tasks simultaneously using multi-task learning. This can save computation at inference time as only a single network needs to be evaluated. Unfortunately, this often leads to inferior overall performance as task objectives can compete, which consequently poses the question: which tasks should and should not be learned together in one network when employing multi-task learning? We study task cooperation and competition in several different learning settings and propose a framework for assigning tasks to a few neural networks such that cooperating tasks are computed by the same neural network, while competing tasks are computed by different networks. Our framework offers a time-accuracy trade-off and can produce better accuracy using less inference time than not only a single large multi-task neural network but also many single-task networks.

A comprehensive semantic understanding of a scene is important for many applications - but in what space should diverse semantic information (e.g., objects, scene categories, material types, 3D shapes, etc.) be grounded and what should be its structure? Aspiring to have one unified structure that hosts diverse types of semantics, we follow the Scene Graph paradigm in 3D, generating a 3D Scene Graph. Given a 3D mesh and registered panoramic images, we construct a graph that spans the entire building and includes semantics on objects (e.g., class, material, shape and other attributes), rooms (e.g., function, illumination type, etc.) and cameras (e.g., location, etc.), as well as the relationships among these entities. However, this process is prohibitively labor heavy if done manually. To alleviate this we devise a semi-automatic framework that employs existing detection methods and enhances them using two main constraints: I. framing of query images sampled on panoramas to maximize the performance of 2D detectors, and II. multi-view consistency enforcement across 2D detections that originate in different camera locations.

Task transfer learning is a popular technique in image processing applications that uses pre-trained models to reduce the supervision cost of related tasks. An important question is to determine task transferability, i.e. given a common input domain, estimating to what extent representations learned from a source task can help in learning a target task. Typically, transferability is either measured experimentally or inferred through task relatedness, which is often defined without a clear operational meaning. In this paper, we present a novel metric, H-score, an easily-computable evaluation function that estimates the performance of transferred representations from one task to another in classification problems using statistical and information theoretic principles. Experiments on real image data show that our metric is not only consistent with the empirical transferability measurement, but also useful to practitioners in applications such as source model selection and task transfer curriculum learning.

Developing visual perception models for active agents and sensorimotor control in the physical world are cumbersome as existing algorithms are too slow to efficiently learn in real-time and robots are fragile and costly. This has given rise to learning-in-simulation which consequently casts a question on whether the results transfer to real-world. In this paper, we investigate developing real-world perception for active agents, propose Gibson Environment for this purpose, and showcase a set of perceptual tasks learned therein. Gibson is based upon virtualizing real spaces, rather than artificially designed ones, and currently includes over 1400 floor spaces from 572 full buildings. The main characteristics of Gibson are: I. being from the real-world and reflecting its semantic complexity, II. having an internal synthesis mechanism "Goggles" enabling deploying the trained models in real-world without needing domain adaptation, III. embodiment of agents and making them subject to constraints of physics and space.

Do visual tasks have relationships, or are they unrelated? For instance, could having surface normals simplify estimating the depth of an image? Intuition answers these questions positively, implying existence of a certain structure among visual tasks. Knowing this structure has notable values; it provides a principled way for identifying relationships across tasks, for instance, in order to reuse supervision among tasks with redundancies or solve many tasks in one system without piling up the complexity. We propose a fully computational approach for modeling the transfer learning structure of the space of visual tasks. This is done via finding transfer learning dependencies across tasks in a dictionary of twenty-six 2D, 2.5D, 3D, and semantic tasks. The product is a computational taxonomic map among tasks for transfer learning, and we exploit it to reduce the demand for labeled data. For example, we show that the total number of labeled datapoints needed for solving a set of 10 tasks can be reduced by roughly 2/3 (compared to training independently) while keeping the performance nearly the same. We provide a set of tools for computing and visualizing this taxonomical structure at http://taskonomy.vision.

Urrently, the most successful learning models in computer vision are based on learning successive representations followed by a decision layer. This is usually actualized through feedforward multilayer neural networks, e.g. ConvNets, where each layer forms one of such successive representations. However, an alternative that can achieve the same goal is a feedback based approach in which the representation is formed in an iterative manner based on a feedback received from previous iterations output. We establish that a feedback based approach has several core advantages over feedforward: it enables making early predictions at the query time, its output naturally conforms to a hierarchical structure in the label space (e.g. a taxonomy), and it provides a new basis for Curriculum Learning. We observe that feedback develops a considerably different representation compared to feedforward counterparts, in line with the aforementioned advantages. We provide a general feedback based learning architecture, instantiated using existing RNNs, with the endpoint results on par or better than existing feedforward networks and the addition of the above advantages.

In this paper, we propose a method for semantic parsing the 3D point cloud of an entire building using a hierarchical approach: first, the raw data is parsed into semantically meaningful spaces (e.g. rooms, etc) that are aligned into a canonical reference coordinate system. Second, the spaces are parsed into their structural and building elements (e.g. walls, columns, etc). Performing these with a strong notation of global 3D space is the backbone of our method. The alignment in the first step injects strong 3D priors from the canonical coordinate system into the second step for discovering elements. This allows diverse challenging scenarios as man-made indoor spaces often show recurrent geometric patterns while the appearance features can change drastically. We also argue that identification of structural elements in indoor spaces is essentially a detection problem, rather than segmentation which is commonly used. We evaluated our method on a new dataset of several buildings with a covered area of over 6, 000m2and over 215 million points, demonstrating robust results readily useful for practical applications.

Deep Recurrent Neural Network architectures, though remarkably capable at modeling sequences, lack an intuitive high-level spatio-temporal structure. That is while many problems in computer vision inherently have an underlying high-level structure and can benefit from it. Spatiotemporal graphs are a popular tool for imposing such high-level intuitions in the formulation of real world problems. In this paper, we propose an approach for combining the power of high-level spatio-temporal graphs and sequence learning success of Recurrent Neural Networks (RNNs). We develop a scalable method for casting an arbitrary spatio-temporal graph as a rich RNN mixture that is feedforward, fully differentiable, and jointly trainable. The proposed method is generic and principled as it can be used for transforming any spatio-temporal graph through employing a certain set of well defined steps. The evaluations of the proposed approach on a diverse set of problems, ranging from modeling human motion to object interactions, shows improvement over the state-of-the-art with a large margin. We expect this method to empower new approaches to problem formulation through high-level spatio-temporal graphs and Recurrent Neural Networks.

Realistic videos of human actions exhibit rich spatiotemporal structures at multiple levels of granularity: an action can always be decomposed into multiple finer-grained elements in both space and time. To capture this intuition, we propose to represent videos by a hierarchy of mid-level action elements (MAEs), where each MAE corresponds to an action-related spatiotemporal segment in the video. We introduce an unsupervised method to generate this representation from videos. Our method is capable of distinguishing action-related segments from background segments and representing actions at multiple spatiotemporal resolutions. Given a set of spatiotemporal segments generated from the training data, we introduce a discriminative clustering algorithm that automatically discovers MAEs at multiple levels of granularity. We develop structured models that capture a rich set of spatial, temporal and hierarchical relations among the segments, where the action label and multiple levels of MAE labels are jointly inferred. The proposed model achieves state-of-the-art performance in multiple action recognition benchmarks. Moreover, we demonstrate the effectiveness of our model in real-world applications such as action recognition in large-scale untrimmed videos and action parsing.

Human communication typically has an underlying structure. This is reflected in the fact that in many user generated videos, a starting point, ending, and certain objective steps between these two can be identified. In this paper, we propose a method for parsing a video into such semantic steps in an unsupervised way. The proposed method is capable of providing a semantic "storyline" of the video composed of its objective steps. We accomplish this utilizing both visual and language cues in a joint generative model. The proposed method can also provide a textual description for each of identified semantic steps and video segments. We evaluate this method on a large number of complex YouTube videos and show results of unprecedented quality for this new and impactful problem.

We address the problem of classifying complex videos based on their content. A typical approach to this problem is performing the classification using semantic attributes, commonly termed concepts, which occur in the video. In this paper, we propose a contextual approach to video classification based on Generalized Maximum Clique Problem (GMCP) which uses the co-occurrence of concepts as the context model. To be more specific, we propose to represent a class based on the co-occurrence of its concepts and classify a video based on matching its semantic co-occurrence pattern to each class representation. We perform the matching using GMCP which finds the strongest clique of co-occurring concepts in a video. We argue that, in principal, the co-occurrence of concepts yields a richer representation of a video compared to most of the current approaches. Additionally, we propose a novel optimal solution to GMCP based on Mixed Binary Integer Programming (MBIP). The evaluations show our approach, which opens new opportunities for further research in this direction, outperforms several well established video classification methods.

The number of GPS-tagged images available on the web is increasing at a rapid rate. The majority of such location tags are specified by the users, either through manual tagging or localization-chips embedded in the cameras. However, a known issue with user shared images is the unreliability of such GPS-tags. In this paper, we propose a method for addressing this problem. We assume a large dataset of GPS-tagged images which includes an unknown subset with contaminated tags is available. We develop a robust method for identification and refinement of this subset using the rest of the images in the dataset. In the proposed method, we form a large number of triplets of matching images and use them for estimating the location of the query image utilizing structure from motion. Some of the generated estimations may be inaccurate due to the noisy GPS-tags in the dataset. Therefore, we perform Random Walks on the estimations in order to identify the subset with the maximal agreement. Finally, we estimate the GPS-tag of the query utilizing the identified consistent subset using a weighted mean. We propose a new damping factor for Random Walks which conforms to the level of noise in the input, and consequently, robustifies Random Walks. We evaluated the proposed framework on a dataset of over 18k user-shared images; the experiments show our method robustly improves the accuracy of GPS-tags under diverse scenarios.

In this paper, we present a new framework for geo-locating an image utilizing a novel multiple nearest neighbor feature matching method using Generalized Minimum Clique Graphs (GMCP). First, we extract local features (e.g., SIFT) from the query image and retrieve a number of nearest neighbors for each query feature from the reference data set. Next, we apply our GMCP-based feature matching to select a single nearest neighbor for each query feature such that all matches are globally consistent. Our approach to feature matching is based on the proposition that the first nearest neighbors are not necessarily the best choices for finding correspondences in image matching. Therefore, the proposed method considers multiple reference nearest neighbors as potential matches and selects the correct ones by enforcing consistency among their global features (e.g., GIST) using GMCP. In this context, we argue that using a robust distance function for finding the similarity between the global features is essential for the cases where the query matches multiple reference images with dissimilar global features. Towards this end, we propose a robust distance function based on the Gaussian Radial Basis Function (G-RBF). We evaluated the proposed framework on a new data set of 102k street view images; the experiments show it outperforms the state of the art by 10 percent.

In this paper we investigate a new problem called visual business recognition. Automatic identification of businesses in images is an interesting task with plenty of potential applications especially for mobile device users. We propose a multimodal approach which incorporates business directories, textual information, and web images in a unified framework. We assume the query image is associated with a coarse location tag and utilize business directories for extracting an over complete list of nearby businesses which may be visible in the image. We use the name of nearby businesses as search keywords in order to automatically collect a set of relevant images from the web and perform image matching between them and the query. Additionally, we employ a text processing method customized for business recognition which is assisted by nearby business names; we fuse the information acquired from image matching and text processing in a probabilistic framework to recognize the businesses. We tested the proposed algorithm on a challenging set of user-uploaded and street view images with promising results for this new application.

This paper presents a novel method for estimating the geospatial trajectory of a moving camera with unknown intrinsic parameters, in a city-scale urban environment. The proposed method is based on a three step process that includes: 1) finding the best visual matches of individual images to a dataset of geo-referenced street view images, 2) Bayesian tracking to estimate the frame localization and its temporal evolution, and 3) a trajectory reconstruction algorithm to eliminate inconsistent estimations. As a result of matching features in query image with the features in the reference geo-taged images, in the first step, we obtain a distribution of geolocated votes of matching features which is interpreted as the likelihood of the location (latitude and longitude) given the current observation. In the second step, Bayesian tracking framework is used to estimate the temporal evolution of frame geolocalization based on the previous state probabilities and current likelihood. Finally, once a trajectory is estimated, we perform a Minimum Spanning Trees (MST) based trajectory reconstruction algorithm to eliminate trajectory loops or noisy estimations. The proposed method was tested on sixty minutes of video, which included footage downloaded from YouTube and footage captured by random users in Orlando and Pittsburgh.