Self-supervised learning (SSL) can be used to solve complex visual tasks without human labels. Self-supervised representations encode useful semantic information about images, and as a result, they have already been used for tasks such as unsupervised semantic segmentation. In this paper, we investigate self-supervised representations for instance segmentation without any manual annotations. We find that the features of different SSL methods vary in their level of instance-awareness. In particular, DINO features, which are known to be excellent semantic descriptors, lack behind MAE features in their sensitivity for separating instances.
3D semantic scene understanding is a fundamental challenge in computer vision. It enables mobile agents to autonomously plan and navigate arbitrary environments. SSC formalizes this challenge as jointly estimating dense geometry and semantic information from sparse observations of a scene. Current methods for SSC are generally trained on 3D ground truth based on aggregated LiDAR scans. This process relies on special sensors and annotation by hand which are costly and do not scale well. To overcome this issue, our work presents the first self-supervised approach to SSC called S4C that does not rely on 3D ground truth data. Our proposed method can reconstruct a scene from a single image and only relies on videos and pseudo segmentation ground truth generated from off-the-shelf image segmentation network during training. Unlike existing methods, which use discrete voxel grids, we represent scenes as implicit semantic fields. This formulation allows querying any point within the camera frustum for occupancy and semantic class. Our architecture is trained through rendering-based self-supervised losses. Nonetheless, our method achieves performance close to fully supervised state-of-the-art methods. Additionally, our method demonstrates strong generalization capabilities and can synthesize accurate segmentation maps for far away viewpoints.
Instructional videos are an excellent source for learning multimodal representations by leveraging video-subtitle pairs extracted with automatic speech recognition systems (ASR) from the audio signal in the videos. However, in contrast to human-annotated captions, both speech and subtitles naturally differ from the visual content of the videos and thus provide only noisy supervision for multimodal learning. As a result, large-scale annotation-free web video training data remains sub-optimal for training text-video models. In this work, we propose to leverage the capability of large language models (LLMs) to obtain fine-grained video descriptions aligned with videos. Specifically, we prompt an LLM to create plausible video descriptions based on ASR narrations of the video for a large-scale instructional video dataset. To this end, we introduce a prompting method that is able to take into account a longer text of subtitles, allowing us to capture context beyond a single sentence. To align the captions to the video temporally, we prompt the LLM to generate timestamps for each produced caption based on the subtitles. In this way, we obtain human-style video captions at scale without human supervision. We apply our method to the subtitles of the HowTo100M dataset, creating a new large-scale dataset, HowToCaption. Our evaluation shows that the resulting captions not only significantly improve the performance over many different benchmark datasets for text-video retrieval but also lead to a disentangling of textual narration from the audio, boosting performance in text-video-audio tasks.
Methods for video motion prediction either estimate jointly the instantaneous motion of all points in a given video frame using optical flow or independently track the motion of individual points throughout the video. The latter is true even for powerful deep-learning methods that can track points through occlusions. Tracking points individually ignores the strong correlation that can exist between the points, for instance, because they belong to the same physical object, potentially harming performance. In this paper, we thus propose CoTracker, an architecture that jointly tracks multiple points throughout an entire video. This architecture combines several ideas from the optical flow and tracking literature in a new, flexible and powerful design. It is based on a transformer network that models the correlation of different points in time via specialised attention layers. The transformer iteratively updates an estimate of several trajectories. It can be applied in a sliding-window manner to very long videos, for which we engineer an unrolled training loop. It can track from one to several points jointly and supports adding new points to track at any time. The result is a flexible and powerful tracking algorithm that outperforms state-of-the-art methods in almost all benchmarks.
Camera pose estimation is a long-standing computer vision problem that to date often relies on classical methods, such as handcrafted keypoint matching, RANSAC and bundle adjustment. In this paper, we propose to formulate the Structure from Motion (SfM) problem inside a probabilistic diffusion framework, modelling the conditional distribution of camera poses given input images. This novel view of an old problem has several advantages. (i) The nature of the diffusion framework mirrors the iterative procedure of bundle adjustment. (ii) The formulation allows a seamless integration of geometric constraints from epipolar geometry. (iii) It excels in typically difficult scenarios such as sparse views with wide baselines. (iv) The method can predict intrinsics and extrinsics for an arbitrary amount of images. We demonstrate that our method PoseDiffusion significantly improves over the classic SfM pipelines and the learned approaches on two real-world datasets. Finally, it is observed that our method can generalize across datasets without further training. Project page: https://posediffusion.github.io/
The variety of objects in the real world is nearly unlimited and is thus impossible to capture using models trained on a fixed set of categories. As a result, in recent years, open-vocabulary methods have attracted the interest of the community. This paper proposes a new method for zero-shot open-vocabulary segmentation. Prior work largely relies on contrastive training using image-text pairs, leveraging grouping mechanisms to learn image features that are both aligned with language and well-localised. This however can introduce ambiguity as the visual appearance of images with similar captions often varies. Instead, we leverage the generative properties of large-scale text-to-image diffusion models to sample a set of support images for a given textual category. This provides a distribution of appearances for a given text circumventing the ambiguity problem. We further propose a mechanism that considers the contextual background of the sampled images to better localise objects and segment the background directly. We show that our method can be used to ground several existing pre-trained self-supervised feature extractors in natural language and provide explainable predictions by mapping back to regions in the support set. Our proposal is training-free, relying on pre-trained components only, yet, shows strong performance on a range of open-vocabulary segmentation benchmarks, obtaining a lead of more than 10% on the Pascal VOC benchmark.
We present Viewset Diffusion: a framework for training image-conditioned 3D generative models from 2D data. Image-conditioned 3D generative models allow us to address the inherent ambiguity in single-view 3D reconstruction. Given one image of an object, there is often more than one possible 3D volume that matches the input image, because a single image never captures all sides of an object. Deterministic models are inherently limited to producing one possible reconstruction and therefore make mistakes in ambiguous settings. Modelling distributions of 3D shapes is challenging because 3D ground truth data is often not available. We propose to solve the issue of data availability by training a diffusion model which jointly denoises a multi-view image set.We constrain the output of Viewset Diffusion models to a single 3D volume per image set, guaranteeing consistent geometry. Training is done through reconstruction losses on renderings, allowing training with only three images per object. Our design of architecture and training scheme allows our model to perform 3D generation and generative, ambiguity-aware single-view reconstruction in a feed-forward manner. Project page: szymanowiczs.github.io/viewset-diffusion.
We consider the problem of reconstructing a dynamic scene observed from a stereo camera. Most existing methods for depth from stereo treat different stereo frames independently, leading to temporally inconsistent depth predictions. Temporal consistency is especially important for immersive AR or VR scenarios, where flickering greatly diminishes the user experience. We propose DynamicStereo, a novel transformer-based architecture to estimate disparity for stereo videos. The network learns to pool information from neighboring frames to improve the temporal consistency of its predictions. Our architecture is designed to process stereo videos efficiently through divided attention layers. We also introduce Dynamic Replica, a new benchmark dataset containing synthetic videos of people and animals in scanned environments, which provides complementary training and evaluation data for dynamic stereo closer to real applications than existing datasets. Training with this dataset further improves the quality of predictions of our proposed DynamicStereo as well as prior methods. Finally, it acts as a benchmark for consistent stereo methods.
We present Farm3D, a method to learn category-specific 3D reconstructors for articulated objects entirely from "free" virtual supervision from a pre-trained 2D diffusion-based image generator. Recent approaches can learn, given a collection of single-view images of an object category, a monocular network to predict the 3D shape, albedo, illumination and viewpoint of any object occurrence. We propose a framework using an image generator like Stable Diffusion to generate virtual training data for learning such a reconstruction network from scratch. Furthermore, we include the diffusion model as a score to further improve learning. The idea is to randomise some aspects of the reconstruction, such as viewpoint and illumination, generating synthetic views of the reconstructed 3D object, and have the 2D network assess the quality of the resulting image, providing feedback to the reconstructor. Different from work based on distillation which produces a single 3D asset for each textual prompt in hours, our approach produces a monocular reconstruction network that can output a controllable 3D asset from a given image, real or generated, in only seconds. Our network can be used for analysis, including monocular reconstruction, or for synthesis, generating articulated assets for real-time applications such as video games.