Learning Zero-Shot Material States Segmentation, by Implanting Natural Image Patterns in Synthetic Data

Visual understanding and segmentation of materials and their states is fundamental for understanding the physical world. The infinite textures, shapes, and often blurry boundaries formed by materials make this task particularly hard to generalize. Whether it's identifying wet regions of a surface, minerals in rocks, infected regions in plants, or pollution in water, each material state has its own unique form. For neural nets to learn general class-agnostic materials segmentation it is necessary to first collect and annotate data that capture this complexity. Collecting and manually annotating real-world images is limited by the cost and precision of manual labor. In contrast, synthetic CGI data is highly accurate and almost cost-free but fails to replicate the vast diversity of the material world. This work offers a method to bridge this crucial gap, by implanting patterns extracted from real-world images, in synthetic data. Hence, patterns automatically collected from natural images are used to map materials into synthetic scenes. This unsupervised approach allows the generated data to capture the vast complexity of the real world while maintaining the precision and scale of synthetic data. We also present the first general benchmark for class-agnostic material state segmentation. The benchmark contains a wide range of real-world images of material states, from cooking, food, rocks, construction, plants, and liquids each in various states (wet/dry/stained/cooked/burned/worn/rusted/sediment/foam...). The annotation includes both partial similarity between regions with similar but not identical materials, and hard segmentation of only points of the exact same material state. We show that net trains on MatSeg significantly outperform existing state-of-the-art methods on this task. The dataset, code, and trained model are available.

One-shot recognition of any material anywhere using contrastive learning with physics-based rendering

We present MatSim: a synthetic dataset, a benchmark, and a method for computer vision based recognition of similarities and transitions between materials and textures, focusing on identifying any material under any conditions using one or a few examples (one-shot learning). The visual recognition of materials is essential to everything from examining food while cooking to inspecting agriculture, chemistry, and industrial products. In this work, we utilize giant repositories used by computer graphics artists to generate a new CGI dataset for material similarity. We use physics-based rendering (PBR) repositories for visual material simulation, assign these materials random 3D objects, and render images with a vast range of backgrounds and illumination conditions (HDRI). We add a gradual transition between materials to support applications with a smooth transition between states (like gradually cooked food). We also render materials inside transparent containers to support beverage and chemistry lab use cases. We then train a contrastive learning network to generate a descriptor that identifies unfamiliar materials using a single image. We also present a new benchmark for a few-shot material recognition that contains a wide range of real-world examples, including the state of a chemical reaction, rotten/fresh fruits, states of food, different types of construction materials, types of ground, and many other use cases involving material states, transitions and subclasses. We show that a network trained on the MatSim synthetic dataset outperforms state-of-the-art models like Clip on the benchmark, despite being tested on material classes that were not seen during training. The dataset, benchmark, code and trained models are available online.

Predicting the Future of AI with AI: High-quality link prediction in an exponentially growing knowledge network

A tool that could suggest new personalized research directions and ideas by taking insights from the scientific literature could significantly accelerate the progress of science. A field that might benefit from such an approach is artificial intelligence (AI) research, where the number of scientific publications has been growing exponentially over the last years, making it challenging for human researchers to keep track of the progress. Here, we use AI techniques to predict the future research directions of AI itself. We develop a new graph-based benchmark based on real-world data -- the Science4Cast benchmark, which aims to predict the future state of an evolving semantic network of AI. For that, we use more than 100,000 research papers and build up a knowledge network with more than 64,000 concept nodes. We then present ten diverse methods to tackle this task, ranging from pure statistical to pure learning methods. Surprisingly, the most powerful methods use a carefully curated set of network features, rather than an end-to-end AI approach. It indicates a great potential that can be unleashed for purely ML approaches without human knowledge. Ultimately, better predictions of new future research directions will be a crucial component of more advanced research suggestion tools.

Seeing Glass: Joint Point Cloud and Depth Completion for Transparent Objects

The basis of many object manipulation algorithms is RGB-D input. Yet, commodity RGB-D sensors can only provide distorted depth maps for a wide range of transparent objects due light refraction and absorption. To tackle the perception challenges posed by transparent objects, we propose TranspareNet, a joint point cloud and depth completion method, with the ability to complete the depth of transparent objects in cluttered and complex scenes, even with partially filled fluid contents within the vessels. To address the shortcomings of existing transparent object data collection schemes in literature, we also propose an automated dataset creation workflow that consists of robot-controlled image collection and vision-based automatic annotation. Through this automated workflow, we created Toronto Transparent Objects Depth Dataset (TODD), which consists of nearly 15000 RGB-D images. Our experimental evaluation demonstrates that TranspareNet outperforms existing state-of-the-art depth completion methods on multiple datasets, including ClearGrasp, and that it also handles cluttered scenes when trained on TODD. Code and dataset will be released at https://www.pair.toronto.edu/TranspareNet/

Predicting 3D shapes, masks, and properties of materials, liquids, and objects inside transparent containers, using the TransProteus CGI dataset

We present TransProteus, a dataset, and methods for predicting the 3D structure, masks, and properties of materials, liquids, and objects inside transparent vessels from a single image without prior knowledge of the image source and camera parameters. Manipulating materials in transparent containers is essential in many fields and depends heavily on vision. This work supplies a new procedurally generated dataset consisting of 50k images of liquids and solid objects inside transparent containers. The image annotations include 3D models, material properties (color/transparency/roughness...), and segmentation masks for the vessel and its content. The synthetic (CGI) part of the dataset was procedurally generated using 13k different objects, 500 different environments (HDRI), and 1450 material textures (PBR) combined with simulated liquids and procedurally generated vessels. In addition, we supply 104 real-world images of objects inside transparent vessels with depth maps of both the vessel and its content. We propose a camera agnostic method that predicts 3D models from an image as an XYZ map. This allows the trained net to predict the 3D model as a map with XYZ coordinates per pixel without prior knowledge of the image source. To calculate the training loss, we use the distance between pairs of points inside the 3D model instead of the absolute XYZ coordinates. This makes the loss function translation invariant. We use this to predict 3D models of vessels and their content from a single image. Finally, we demonstrate a net that uses a single image to predict the material properties of the vessel content and surface.

Computer vision for liquid samples in hospitals and medical labs using hierarchical image segmentation and relations prediction

This work explores the use of computer vision for image segmentation and classification of medical fluid samples in transparent containers (for example, tubes, syringes, infusion bags). Handling fluids such as infusion fluids, blood, and urine samples is a significant part of the work carried out in medical labs and hospitals. The ability to accurately identify and segment the liquids and the vessels that contain them from images can help in automating such processes. Modern computer vision typically involves training deep neural nets on large datasets of annotated images. This work presents a new dataset containing 1,300 annotated images of medical samples involving vessels containing liquids and solid material. The images are annotated with the type of liquid (e.g., blood, urine), the phase of the material (e.g., liquid, solid, foam, suspension), the type of vessel (e.g., syringe, tube, cup, infusion bottle/bag), and the properties of the vessel (transparent, opaque). In addition, vessel parts such as corks, labels, spikes, and valves are annotated. Relations and hierarchies between vessels and materials are also annotated, such as which vessel contains which material or which vessels are linked or contain each other. Three neural networks are trained on the dataset: One network learns to detect vessels, a second net detects the materials and parts inside each vessel, and a third net identifies relationships and connectivity between vessels.

Deep Molecular Dreaming: Inverse machine learning for de-novo molecular design and interpretability with surjective representations

Computer-based de-novo design of functional molecules is one of the most prominent challenges in cheminformatics today. As a result, generative and evolutionary inverse designs from the field of artificial intelligence have emerged at a rapid pace, with aims to optimize molecules for a particular chemical property. These models 'indirectly' explore the chemical space; by learning latent spaces, policies, distributions or by applying mutations on populations of molecules. However, the recent development of the SELFIES string representation of molecules, a surjective alternative to SMILES, have made possible other potential techniques. Based on SELFIES, we therefore propose PASITHEA, a direct gradient-based molecule optimization that applies inceptionism techniques from computer vision. PASITHEA exploits the use of gradients by directly reversing the learning process of a neural network, which is trained to predict real-valued chemical properties. Effectively, this forms an inverse regression model, which is capable of generating molecular variants optimized for a certain property. Although our results are preliminary, we observe a shift in distribution of a chosen property during inverse-training, a clear indication of PASITHEA's viability. A striking property of inceptionism is that we can directly probe the model's understanding of the chemical space it was trained on. We expect that extending PASITHEA to larger datasets, molecules and more complex properties will lead to advances in the design of new functional molecules as well as the interpretation and explanation of machine learning models.

Generator evaluator-selector net: a modular approach for panoptic segmentation

In machine learning and other fields, suggesting a good solution to a problem is usually a harder task than evaluating the quality of such a solution. This asymmetry is the basis for a large number of selection oriented methods that use a generator system to guess a set of solutions and an evaluator system to rank and select the best solutions. This work examines the use of this approach to the problem of image segmentation. The generator/evaluator approach for this case consists of two independent convolutional neural nets: a generator net that suggests variety segments corresponding to objects and distinct regions in the image and an evaluator net that chooses the best segments to be merged into the segmentation map. The result is a trial and error evolutionary approach in which a generator that guesses segments with low average accuracy, but with wide variability, can still produce good results when coupled with an accurate evaluator. Generating and evaluating each segment separately is essential in this case since it demands exponentially fewer guesses compared to a system that guesses and evaluates the full segmentation map in each try. Another form of modularity used in this system is separating the segmentation and classification into independent neural nets. This allows the segmentation to be class agnostic and hence capable of segmenting unfamiliar categories that were not part of the training set. The method was examined on the COCO Panoptic segmentation benchmark and gave competitive results to the standard semantic segmentation and instance segmentation methods.