For the state of the art for segmentation on the Cityscapes dataset, PSPNet isnt there and to my knowledge at 2017, it was state of the art (SOTA). I dont see it on this site.
Why is Facebook getting into reinforcement learning? Are there applications of RL to improving social connections or extracting more data to sell to advertisers?
I am super excited about this work and network dissection. The idea of causal units or finding neurons that correlate to objects in an image will have huge impact on image manipulation tasks. I am excited to try this to control GANS on how they generate images.
Isn't using pretrained CNN models make embedding biased towards extracting features that were relevant from the dataset (i.e. ImageNet) it learned from? Fine-grained classification uses triplet and siamese networks to learn an embedding based on semantic similarity, but you have to define what is semantically similar in the dataset. I am curious how well pretrain networks generalize for indexing for image search. I think finding papers how pinterest and ebay apply visual search at scale may shed some light on this.
I wish there are more articles like this. Seeing how performance characteristics (specificity/sensitivity) effect business outcomes is really insightful! It would be cool if you can improve the performance metric of this model, how much money/impact that would make on a business.
Are there any good guides, tutorials, or research papers that investigate or advise how to inspect weights during training for debugging. The only things I read are to watch out for vanishing gradients, and when fine-tuning the most change in layers are seen toward the end of the network, not the beginning layers.
Software development is hard, but why does it have to be taught at a university, where you learn core concepts but are not learning skills and experiences that companies are looking for? I think to address your concerns, if Lambda can have continuing education where you can take more abstract and CS classes, students can converge to have the same knowledge as a CS student and be more skilled to integrate that knowledge.
Snapchat’s lens studio product is a definite category defining product. As an AR developer, it’s been soo much fun and enjoyable creating AR experiences, and its one of the first times I can make an experience and directly engage a large audience.
I have been collecting similar guides to improve on software engineering, as I am a recent graduate looking to excel in my career. Here are some other blogs I found useful:
Here is a demo of a CoreML model I develop to recognize hand signs, running in realtime on iOS camera output. The architecture is a MobileNet v1 classification architecture. The model is 13 Megabytes in size, and the average processing time is 40ms. Accuracy on Dev and test set is 94%!
Dataset location and inspiration was a tutorial from this link:
The demo here is running on an iPhone 6s. The model was architected using Keras, trained on CS230's SIGNS dataset and custom dataset, and converted to coreml using coremltools.
I don't see how this article addresses the person's question that if you want to work in machine learning, should you do a PhD? From my experience, companies want a few years(3+) of software engineering experience and a graduate degree of some sort, whether a masters or PhD. From my experience after getting a masters, Ive gone through the hassle of interviewing and be turned down bec they are looking for more experienced people for ML Eng and I only had 1 year of professional eng experience.
Looks interesting! What are common use cases to use this kind of library? I was thinking of using it to automate uptime of servers or exploring adding NLP to emails, but I am interested what are popular uses for this kind of a framework.
You should email employees and engineers from Aut Veh startups and see what they did or what they are looking for. The udacity Autonomous nanodegree gives an idea of what skills are relevant, but you have to see/learn what skills are they looking for enough for them to actually hire you. I completed a masters degree from a good university and did some research in machine learning. I think an advanced degree helps with credibility as well.
Exciting to see this! I can attest to how programs like this can be so impactful. I attended a summer research program during my undergrad called the MIT Summer Research program, and it changed my life. The program enabled me to work and get connected with some of the most brilliant professors and researchers in the world. I worked at the MIT Media Lab, where it instilled my passion for Augmented Reality and software engineering. I leveraged that internship to get letters of recommendations, which enabled me to be accepted to an Ivy league school for graduate school, a dream that I didn't think was possible for me.
How easy is it to build custom architectures like segmentation and object detection networks? I am working on a segmentation coreml tools and apple coremltools only coverts a selection of layers.
I wonder what the details of the computer vision tracking are. I am thinking its simple color tracking which has challenges under lighting and contrast with other colors. Object detection using deep learning could make the tracking more robust.
Hi there! I recently graduated with my masters from Cornell and am really interested in the NLP junior software engineer position. Is there an email to a hiring manager or a team member you can share that I can contact directly?
"Pick nodes on the knowledge graph which are highly-connected to things you do routinely". I am really interested in knowing what are the nodes I should focus on for data science and machine learning? My thoughts are first focusing on the tools and libraries I interact with every day, then the harder/high level concepts.
Creator here! I am really excited to share this Augmented Reality experience and see what you all think about it. The idea of that the camera is your paint brush, and the world is your canvas really stuck out to me, so I made this so people can doodle and sketch in your world! Let me know what you all think and what other features I should add.
I’m sooo impressed with the quality of the models! Have you considered showcasing using farious AR platforms, like ARKit or snapchat’s Lens studio? I am developing stuff on lens studio and I think it’s a great way to advertise!
This is amazing, I would love to something similar to this in Augmented Reality, here is a snapchat AR experience where you visualize several waveforms: