Winter Conference on Applications of Computer Vision 2026 - Monday WACV DAILY
14:45-15:45 OMeGa: Joint Optimization of Explicit Meshes and Gaussian Splats … 14:45-15:45 Occlusion Boundary and Depth: Mutual Enhancement via MultiTask … 14:45-15:45 Broadcast2Pitch: Game State Reconstruction from Unconstrained … Poster Session 3-18: Large Sign Language Models: Toward 3D American Sign … Poster Session 3-93: Gen-AFFECT: Generation of Avatar Fine-grained Facial … Poster Session 4-13: Spec-Gloss Surfels and Normal-Diffuse Priors for Relightable ... Jorge’s picks of the day: Jorge Lopez-Moreno (first from left) is a professor and head of the Multimodal Simulation Lab (MSLab.es) at Universidad Rey Juan Carlos in Madrid, Spain. For today, Monday 9 2 Jorge’s Picks DAILY WACV Monday Posters “My research sits at the intersection of mechanical simulation and rendering. I work on developing techniques to replicate and edit the appearance of real-world objects - and people - including their geometry and material optical properties. While most of my work is rooted in rendering, it often combines ideas from computer vision and machine learning. I’m particularly excited about real-time applications, and about building tools that support fashion and art design as well as visual effects. One of the best parts of my job is exploring these ideas together with my PhD students and seeing their creativity turn into new projects.” “This year at WACV I’m attending with three of our students - Henar, Pablo, and Mario. Today we presented two projects: AutoSew, a method to automatically stitch 2D patterns into a garment developed with Adobe Systems, and PHYSPLAT, a framework for photorealistic hybrid simulation of real and synthetic elements using 3D Gaussian Splatting, in collaboration with the Arquimea Research Center.” “When I’m not doing research, I enjoy martial arts, painting, and playing videogames with my kids - although they usually end up beating me.” Orals
3 DAILY WACV Monday UKRAINE CORNER Russian Invasion of Ukraine Our sister conference CVPR condemns in the strongest possible terms the actions of the Russian Federation government in invading the sovereign state of Ukraine and engaging in war against the Ukrainian people. We express our solidarity and support for the people of Ukraine and for all those who have been adversely affected by this war. WACV Daily Editor: Ralph Anzarouth Publisher & Copyright: Computer Vision News All rights reserved. Unauthorized reproduction is strictly forbidden. Our editorial choices are fully independent from IEEE, WACV and the conference organizers. Good Morning Tucson! Howdy? Enjoy the reading of this WACV Daily and have a great Monday! Ralph Anzarouth Editor, Computer Vision News Ralph’s photo above was taken in peaceful, lovely and brave Odessa, Ukraine.
When we take a picture in the real world, the image is always constrained by the lens. If part of a person’s face falls outside the frame, the camera cannot capture it. Existing object detection systems are designed around that limitation, so that if something is not visible, they do not detect it. “But the real object will not just disappear by going out of the lens,” Changlin rightly points out. “There are still potential applications and value in detecting objects outside the frame.” That idea became the motivation for this project. The question first came from his supervisor, Dylan 4 DAILY WACV Monday Oral Presentation Changlin Song is a first-year PhD student at the Australian National University. His paper explores a question that most object detection systems avoid: what happens to an object when it moves outside the camera’s field of view? Changlin tells us how a conversation with his supervisor led him to rethink what detection really means. Extreme Amodal Face Detection
5 DAILY WACV Monday Changlin Song Campbell. “He asked me, what can we do with an object that is out of frame?” Changlin recalls. “There are so many works that are trying to improve the accuracy of things within the frame, but there’s not much we can do about things outside.” One solution would be to use generative models to expand the image and then perform detection. “Generative models are definitely a good way,” Changlin says, “but it’s very slow.” For applications where detection must occur instantaneously, particularly in safety-related contexts, that latency is a serious limitation. That led him to a different line of thinking. “Actually, we don’t need the pixel information,” he explains. “Generative models will produce that pixel information, so can we try a lightweight method that only focuses on the things that we care about?” The technical challenge quickly became clear. Expanding the frame in a model’s latent space substantially increases the region’s size. “It’s like exponentially expanding the region we are going to detect,” Changlin tells us. “You are predicting a lot from a little.” At the same time, faces, like many objects, are sparsely distributed. Searching the entire expanded region at full resolution would introduce heavy computational overhead. The team therefore designed
6 DAILY WACV Monday Oral Presentation a selective coarse-to-fine decoder approach: first examining the expanded area at low resolution, then refining only the most promising candidate regions. The result is a lightweight detector that can run at around three to five frames per second. “Basically, you can do it in real time,” Changlin adds. “I guess that’s the biggest value that we’re exploring.” For Changlin, the significance of the work goes beyond performance metrics. “The ultimate goal of computer vision is to make the machine look like a human,” he says. As humans, we do not assume objects vanish when they move beyond our immediate view. We infer their continued presence. “By equipping the machine learning models with this ability, we can make them more reliable, more powerful, and more like a human.”
The project aligns closely with his broader PhD research. “Currently, I’m working on exploring the world’s unseen regions,” he explains. “What we can infer from the scene to detect or infer the unseen world.” Extreme amodal face detection is one step in that direction. When asked about his future plans, Changlin is open-minded. “It kind of depends on what offers I get,” he says with a smile. Industry appeals to him because it offers the chance to build products that contribute directly to the world, but academia remains his focus for now. “Research is the kind of thing that I’m enjoying.” He chose the Australian National University in part because of its strong tradition in computer vision. “We have a really famous professor called Richard Hartley,” he notes, referencing the author of Multiple View Geometry in Computer Vision. “Also, they have a lot of good faculty and good students here.” You can learn more about Changlin’s work during Oral Session 4A: Image Recognition and Understanding II (AZ Ballroom 6) on Monday from 9:45–10:45, and at Poster Session 3 (Tucson Ballroom & Prefunction Space) on Monday from 10:45–12:30, poster #2. 7 DAILY WACV Monday Changlin Song
8 DAILY WACV Monday Oral Presentation Autonomous vehicles typically have an heterogeneous sensor suite installed on board. This is because you need a lot of sensors of different types to enable an intelligent vehicle to drive in an automatic way. A fundamental in this scenario is to fuse all the information that you acquire with your sensors. In order to have consistent and coherent data, you need to know the relations that you have between different sensors. CalibBEV: LiDAR-Camera Calibration via BEV Alignment Filippo D’Addeo (left) is a third year PhD student in Automotive Engineering for Intelligent Mobility at the University of Bologna. Lorenzo Cipelli (right) is a second year PhD student at the University of Parma. Their research activity takes place in Italy at VisLab, an Ambarella company developing autonomous driving vehicle algorithms. Their paper focuses on LiDAR and camera calibration. Both Filippo and Lorenzo are at their first computer vision conference and they start very strong with an oral presentation. Kudos! They talk to us ahead of their oral and poster presentation today.
9 DAILY WACV Monday Filippo D’Addeo - Lorenzo Cipelli Lorenzo explains that all the methods trying to address the calibration between sensors focus on initial classification and then iterative refinement of the alignment of the two modalities. These gave mediocre results and very slow algorithms. Subsequent methods focused on trying to get features between the two modalities as close as possible. “In our method,” he adds “we try to follow this path but in a different way, meaning that we use also geometrical information which is inside or let's say intrinsic into a bird eye view representation, meaning an image of the point cloud from above. So we decided to make features closer between the two modalities and also leverage the geometrical aspect of bird eye view representations.” Then Lorenzo tell us of the challenges that they encountered on the way: “What I remember from our days in the office was that it is really a delicate task, meaning you change something and it could be that you don't get the outcome that you expected and it is not consistent many times.” They solved this with a lot of attention to the details. In some cases, Filippo adds, that meant writing the algorithms on a whiteboard, check for everything, visualize and see if they were right or wrong. They declare that, fortunately, they achieved robust results, meaning that the error is not deviating a lot from the mean of their results. They decided to start their work not from other calibration works but take instead inspiration from another task, which is the object detection task. Many people are more familiar with detection instead of calibration. The idea was to somehow adapt something made for detection, 3D detection, to the calibration task. They did prove in their work that this is possible, obtaining also better results compared to previous calibration methods.
Right now, the direction in the autonomous driving field is going into an end-to-end pipeline, meaning you get data from the sensor, you try to process it, maybe you do it in an early stage, you fuse the data right away when it goes out of the sensor. You can also do it in a later stage, meaning that you process the data from different sensors independently and then you fuse it. “This depends on how you approach the thing,” Lorenzo declares, “but you get the data, you process it and then after fusion you specialize on the task you want. The trend now is going end-to-end. So from the data to the output, which is the action of physical driving.” Filippo thinks that this work solves the problem on one side and it opens new directions for new research on the other side: “I think both are true - we solve the problem by demonstrating that our method performs better than previous methods. Our model’s generalization ability is such that you can train your model with one camera and one LiDAR and then during the evaluation stage you can use another camera, a different camera. And things keep working! But we also open a new direction, because no one before us has focused on the bird's eye view technique for calibration.” 10 DAILY WACV Monday Oral Presentation
A curiosity we discovered during the interview: the three of us are from Emilia, a lovely region in the north of Italy, known for great food and racing cars. If that has piqued your interest, you can learn more about this work during Oral Session 5B: Remote Sensing and Sensors, Monday 13:30–14:30 in AZ Ballroom 7, and during Poster Session 2, Sunday 16:00–17:45 in the Tucson Ballroom and Prefunction Space. 11 DAILY WACV Monday Filippo D’Addeo - Lorenzo Cipelli These women were competitive gymnasts. They became computer vision scientists. They are Italian. They are in tomorrow’s WACV Daily. Meet the scientist behind the science!!!
Lea Bogensperger is a postdoctoral researcher at the University of Zurich in Switzerland. Lea, what is your work about? I'm coming from a PhD in computer science. I worked mainly on inverse problems and image reconstruction, basically tackling different problems in the image acquisition pipeline from alignment and reconstruction to bi-level optimization and then also to segmentation. It was nice for me to see like a broad span in the vision pipeline. And now in my postdoc I transitioned a bit to work also on different kind of medical or biological data. For example, also proteins now but of course I’m still working with optimization and machine learning techniques. Is that what you wanted to do? What I always wanted to do is have some medical applications. I'm still also working on images, medical images from the hospital for example, but that is the most important thing to me. Do you have anything in your work experience that you can already claim as something that you are proud of? It's not near done yet but if that ever happens and I can contribute to that, I would be very proud of. It's an ongoing project. You have a paper here at WACW. I'm a co-author, yes, with a collaborator from Austria who's doing her PhD right now. Of all the things that you have done during your PhD, which one is the most useful for your current work? I think all of them helped me of course to develop my knowledge in the field. But I liked a lot my last project, where I worked on generative methods for image segmentation. I was segmenting microscopy images using diffusion models, so you can have it in a sort of probabilistic way. And I think the methods I learned there when I started my postdoc here in Zurich, I was kind of thrown into a completely different field of applications like protein sequences. But whatever I learned during my PhD on the images I could carry on so much to the proteins now. So I think methodologically you can have a lot of related things but then of course the challenge is to work with new data each time. Was your PhD worth doing? Oh of course, of course. In retrospect you will always say you'd Read 160 FASCINATING interviews with Women in Computer Vision Read 160 FASCINATING interviews with Women in Computer Vision Women in Computer Vision 12 DAILY WACV Monday
13 DAILY WACV Monday Lea Bogensperger
do it again - or most of the times - but I really enjoyed the overall experience. Have you always thought like this during your PhD or there were ups and downs? Oh, there were a lot of ups and downs. Certainly! I think I had like my first major crisis after one and a half years or so when I thought I didn't make any progress. What am I doing here? What really helped me was actually to talk to my professor during that time. I really told him I think it's not what I'm supposed to do and then he was really taking out a lot of the pressure. So that was really good actually to be transparent and he was like you don't have to have a paper published at this point. I see you're on track. Don't worry. That was really good. He was super nice. It helped me a lot. Of course, also later I had ups and downs and also now still in my postdoc. They keep reappearing. I think it's somehow normal. You probably were not tempted enough to really quit. I said that from time to time. Still now maybe sometimes I will think maybe I should look for a job in industry. But then I get back to this point where I know that I love it so much, so I'll just stick to it. Is academia the place of your future or you don't know where you will be in 10 years from now? I don't know. I tried to keep an open mindset to have all possibilities open. but I would really love to stay somehow in this field - you can call it applied research of like ML in health. I think academia is a very cool place to be in general. And I don't think that research necessarily has to be useful by definition. I mean, it is research, right. It is also about learning new things. So I think it doesn't have to be useful in a specific way. I think that's the beautiful thing also about science. But I think it's also cool if all of this great research that is being done is also taken to some practical applications. In the end we need both. It's also cool that there are people that say maybe my stuff is useless but I just love every aspect about it. And then there are other people that say: OK, now I'm focusing on translational research. I think we need it all. Can you think at one wrong thing that you did and you would advise your younger self or our youngest readers not to do the same mistake? It's a good question. I just moved one and a half years to Switzerland Women in Computer Vision 14 DAILY WACV Monday
15 DAILY WACV Monday Lea Bogensperger
Women in Computer Vision 16 DAILY WACV Monday and before that I was always in my hometown same university for a bachelor master and PhD. I think I would have done that step sooner. To immerse my… I mean Switzerland is not far from Austria but I managed at least a tiny step. I think exploring that more. Where are you from in Austria? Graz. Graz university is excellent. Oh, it was amazing. I did enjoy it a lot. You took the choice of spending the best years of your life in Switzerland. What do you miss of Austria? What would you have if you were still in Graz? They are quite similar. I chose to stay in the Alps because I like mountainous stuff a lot. So that is actually pretty similar. I think the only thing that is far away now in Austria for me is my friends and family. But you know, they are one night train away from Zurich. So it's not something I can really complain about.
17 DAILY WACV Monday Lea Bogensperger
Of all the things that you have learned during all these years. Is there one thing that you could teach me on the spot that I could say about: oh, I'm happy to know! That's a very good question. Thank you. Let me think. This is now super hard for me to say because it's very open question. What I mentioned before like generative modeling diffusion models. I'm not going to go into the method in detail but it was super nice for me to see a method that you can use to synthesize new images by learning the distribution. I could use it for segmenting images, like getting the cell shapes out of medical images. But now I can use the same method with data processing to generate new variants Women in Computer Vision 18 DAILY WACV Monday “if you find something that you enjoy, keep doing it; if you have the opportunity and go for possibilities that come up, even though it might not seem you know very clearly from the beginning”
of a protein, even though you know it's a super different application area! But the method in itself, like these lines of code I take from one project to another, they're the same! I like a lot that this this method can be translated. But how can you turn a probabilistic method into something that has a connection to truth and to the real world. Yeah that's a good question, because we're learning with a distribution of known proteins. This is also a bit where I see the limitations if you do purely in silico modeling. We can apply some filters, like computational filters, to check does this fulfill certain criteria. But ultimately, we would have to go to the lab. So there of course we would need the collaboration with wet lab people to synthesize the actual protein. But I think the strength would be of course the use all of these computational methods. And while from a probabilistic method we can generate thousands of samples, we can narrow down the samples to go into testing at the end by using a lot of filters in that pipeline. I guess that's where we also have the strength of our methods. It's not the ultimate truth. I get your question, but we can get a very good set of candidates. That is the hope of course. That's a lovely answer. I love it. Your last words. I guess what I would take away from my past years is if you find something that you enjoy, keep doing it; if you have the opportunity and go for possibilities that come up, even though it might not seem you know very clearly from the beginning, in the end I think you can get to some very cool places! So passion will drive you to cool places. I would say so! 19 DAILY WACV Monday Read 160 FASCINATING interviews with Women in Computer Vision! Read 160 FASCINATING interviews with Women in Computer Vision! Lea Bogensperger
20 DAILY WACV Monday Workshops Above, Matthew Stamm speaking on Friday at SAFE 2026 - Synthetic & Adversarial ForEnsics workshop. Xavi Giró is an applied scientist at Amazon Barcelona, working on image generation. He showed how to cut GenAI review costs with automatic quality assessment, even if it is imperfect. The poster was presented at the 5th Workshop on Image/Video/Audio Quality Assessment in Computer Vision, VLM and Diffusion Model.
Double-DIP Don’t miss the BEST OF WACV 2026 iSCnul i Ccbkos cmhreipbruet feor rVf ri sei eo na nNde wg es toi tf iAnpyr oi l .u r m a i l b o x ! Don’t miss the BEST OF WACV 2026 in Computer Vision News of April. Subscribe for free and get it in your mailbox! Click here Target with solid fill
22 DAILY WACV Congrats, Doctor Silvia! Monday Silvia Seidlitz recently obtained her PhD at Heidelberg University (Germany) under the supervision of Lena Maier-Hein. Her research focused on developing AI-driven hyperspectral imaging for perioperative care and addressing real-world challenges in translating such models into clinical practice. Silvia is now a researcher at Carl Zeiss Semiconductor Manufacturing Technology (Oberkochen, Germany). Congrats, Doctor Silvia! Modern surgery is astonishingly advanced: Robot-assisted systems, realtime imaging and precision instruments are routine. Yet fundamental challenges remain. For instance, when assessing tissue viability, surgeons still largely rely on what looks “pink enough”, guided by experience. However, misjudgements during surgery can lead to postoperative complications, which may progress to infection and even sepsis. Postoperative complications account for 7.7% of global deaths, making them the third leading cause of death. Silvia’s research aimed to improve perioperative care by providing physicians with missing information. She focused on three key clinical applications:
Functional tissue monitoring: Intraoperative estimation of tissue parameters such as oxygenation, enabling the objective identification of perfused versus ischemic tissue. Tissue differentiation: Since distinguishing tissue types can be difficult — even for experienced surgeons — her work explored automated semantic segmentation of surgical scenes. Sepsis diagnosis: As diagnosing postoperative sepsis is often slow and inconclusive and mortality risk increases with every hour of delayed treatment, Silvia investigated rapid and reliable detection of sepsis in postoperative intensive care unit (ICU) patients. To address these challenges, Silvia used hyperspectral imaging (HSI), a technology originally developed for remote sensing. Changes in tissue composition, such as differences in perfusion, tissue type, or early signs of sepsis (e.g., edema and microcirculatory dysfunction), alter the spectral signatures of biological tissue. However, these changes often remain invisible to the human eye. This limitation arises because human vision (and conventional RGB cameras that mimic it) captures only three broad color channels: red, green, and blue. HSI overcomes this restriction by measuring light in many narrow wavelength bands, often extending beyond the visible spectrum, and thus capturing subtle reflectance differences at every pixel. Because HSI data is high dimensional, deep learning is well suited to identify patterns related to tissue physiology and disease. Silvia demonstrated that HSI-based deep learning models for surgical scene segmentation can achieve performance comparable to a human expert. In what is currently the largest clinical HSI study, she further showed that deep learning models can accurately distinguish septic from non-septic patients in a surgical ICU and outperform widely used biomarkers and clinical scores, while enabling rapid, non-invasive, cost-effective and mobile assessments. A major barrier to the clinical deployment of AI systems is domain shift, which can severely degrade model performance. While increasingly studied in computer vision, this problem has received little attention in medical HSI. Silvia addressed this gap by systematically analyzing real-world domain shifts related to hardware, scene geometries and populations. To support further research and clinical translation in this emerging field, Silvia also released public datasets as well as code and pretrained models. Her full thesis is available here. 23 DAILY WACV Monday Silvia Seidlitz
RkJQdWJsaXNoZXIy NTc3NzU=