Mihai Pîrvu

The computer vision domain has greatly benefited from an abundance of data across many modalities to improve on various visual tasks. Recently, there has been a lot of focus on self-supervised pre-training methods through Masked Autoencoders (MAE) \cite{he2022masked,bachmann2022multimae}, usually used as a first step before optimizing for a downstream task, such as classification or regression. This is very useful as it doesn’t require any manually labeled data. In this work, we introduce Probabilistic Hyper-Graphs using Masked Autoencoders (PHG-MAE): a novel model that unifies the classical work on neural graphs \cite{leordeanu2021semi} with the modern approach of masked autoencoders under a common theoretical framework. Through random masking of entire modalities, not just patches, the model samples from the distribution of hyper-edges on each forward pass. Additionally, the model adapts the standard MAE algorithm by combining pre-training and fine-tuning into a single training loop. Moreover, our approach enables the creation of inference-time ensembles which, through aggregation, boost the final prediction performance and consistency. Lastly, we show that we can apply knowledge distillation on top of the ensembles with little loss in performance, even with models that have fewer than 1M parameters. While our work mostly focuses on outdoor UAV scenes that contain multiple world interpretations and modalities, the same steps can be followed in other similar domains, such as autonomous driving or indoor robotics. In order to streamline the process of integrating external pre-trained experts for computer vision multi-modal multi-task learning (MTL) scenarios, we developed a data-pipeline software. Using this tool, we have created and released a fully-automated extension of the Dronescapes dataset. All the technical details, code and reproduction steps are publicly released.

The real-world is inherently multi-modal at its core. Our tools observe and take snapshots of it, in digital form, such as videos or sounds, however much of it is lost. Similarly for actions and information passing between humans, languages are used as a written form of communication. Traditionally, Machine Learning models have been unimodal (i.e. rgb -> semantic or text -> sentiment_class). Recent trends go towards bi-modality, where images and text are learned together, however, in order to truly understand the world, we need to integrate all these independent modalities. In this work we try to combine as many visual modalities as we can using little to no human supervision. In order to do this, we use pre-trained experts and procedural combinations between them on top of raw videos using a fully autonomous data-pipeline, which we also open-source. We then make use of PHG-MAE, a model specifically designed to leverage multi-modal data. We show that this model which was efficiently distilled into a low-parameter (<1M) can have competitive results compared to models of ~300M parameters. We deploy this model and analyze the use-case of real-time semantic segmentation from handheld devices or webcams on commodity hardware. Finally, we deploy other off-the-shelf models using the same framework, such as DPT for near real-time depth estimation.

There are many ways of interpreting the world and they are highly interdependent. We exploit such complex dependencies and introduce a powerful multi-task hypergraph, in which every node is a task and different paths through the hypergraph reaching a given task become unsupervised teachers, by forming ensembles that learn to generate reliable pseudolabels for that task. Each hyperedge is part of an ensemble teacher for a given task and it is also a student of the self-supervised hypergraph system. We apply our model to one of the most important problems of our times, that of Earth Observation, which is highly multi-task and it often suffers from missing ground-truth data. By performing extensive experiments on the NASA NEO Dataset, spanning a period of 22 years, we demonstrate the value of our multi-task semi-supervised approach, by consistent improvements over strong baselines and recent work. We also show that the hypergraph can adapt unsupervised to gradual data distribution shifts and reliably recover, through its multi-task self-supervision process, the missing data for several observational layers for up to seven years.

We present a method for learning multiple scene representations given a small labeled set, by exploiting the relationships between such representations in the form of a multi-task hypergraph. We also show how we can use the hypergraph to improve a powerful pretrained VisTransformer model without any additional labeled data. In our hypergraph, each node is an interpretation layer (e.g., depth or segmentation) of the scene. Within each hyperedge, one or several input nodes predict the layer at the output node. Thus, each node could be an input node in some hyperedges and an output node in others. In this way, multiple paths can reach the same node, to form ensembles from which we obtain robust pseudolabels, which allow self-supervised learning in the hypergraph. We test different ensemble models and different types of hyperedges and show superior performance to other multi-task graph models in the field. We also introduce Dronescapes, a large video dataset captured with UAVs in different complex real-world scenes, with multiple representations, suitable for multi-task learning.

Estimating precise metric depth is an essential task for UAV navigation, which is very difficult to learn unsupervised without access to odometry. At the same time, depth recovery from kinematics and optical flow is mathematically precise, but less numerically stable and robust, especially in the focus of expansion areas. We propose a model that combines the analytical, vision-with-odometry approach, with deep unsupervised learning, into a single formulation for metric depth estimation, which is both fast and accurate. The two pathways – analytical and data-driven – form a robust ensemble, which provides supervisory signal to a single deep net that distills the consensus between scene geometry, pose, kinematics, camera intrinsics and the input RGB. The distilled net has low runtime and memory costs, being suitable for embedded devices. We validate our results against an off-the-shelf SfM-based solution. We also introduce a new real-world dataset of almost 20 minutes of continuous UAV flight, on which we demonstrate better accuracy and capabilities than the deep learning and analytical approaches.

We address the challenging problem of semi-supervised learning in the context of multiple visual interpretations of the world by finding consensus in a graph of neural networks. Each graph node is a scene interpretation layer, while each edge is a deep net that transforms one layer at one node into another from a different node. During the supervised phase edge networks are trained independently. During the next unsupervised stage edge nets are trained on the pseudo-ground truth provided by consensus among multiple paths that reach the nets’ start and end nodes. These paths act as ensemble teachers for any given edge and strong consensus is used for high-confidence supervisory signal. The unsupervised learning process is repeated over several generations, in which each edge becomes a “student” and also part of different ensemble “teachers” for training other students. By optimizing such consensus between different paths, the graph reaches consistency and robustness over multiple interpretations and generations, in the face of unknown labels. We give theoretical justifications of the proposed idea and validate it on a large dataset. We show how prediction of different representations such as depth, semantic segmentation, surface normals and pose from RGB input could be effectively learned through self-supervised consensus in our graph. We also compare to state-of-the-art methods for multi-task and semi-supervised learning and show superior performance.

We propose a method at the intersection of Computer Vision and Computer Graphics fields, which automatically generates RGBD images using neural networks, based on previously seen and synchronized video, depth and pose signals. Since the models must be able to reconstruct both texture (RGB) and structure (Depth), it creates an implicit representation of the scene, as opposed to explicit ones, such as meshes or point clouds. The process can be thought of as neural rendering, where we obtain a function f : Pose -> RGBD, which we can use to navigate through the generated scene, similarly to graphics simulations. We introduce two new datasets, one based on synthetic data with full ground truth information, while the other one being recorded from a drone flight in an university campus, using only video and GPS signals. Finally, we propose a fully unsupervised method of generating datasets from videos alone, in order to train the Pose2RGBD networks. Code and datasets are available at:: https://gitlab.com/mihaicristianpirvu/pose2rgbd.

The emergence of relatively low cost UAVs has prompted a global concern about the safe operation of such devices. Since most of them can ‘autonomously’ fly by means of GPS way-points, the lack of a higher logic for emergency scenarios leads to an abundance of incidents involving property or personal injury. In order to tackle this problem, we propose a small, embeddable ConvNet for both depth and safe landing area estimation. Furthermore, since labeled training data in the 3D aerial field is scarce and ground images are unsuitable, we capture a novel synthetic aerial 3D dataset obtained from 3D reconstructions. We use the synthetic data to learn to estimate depth from in-flight images and segment them into ‘safe-landing’ and ‘obstacle’ regions. Our experiments demonstrate compelling results in practice on both synthetic data and real RGB drone footage.