OpenCV is offering 25% discount on all its official AI courses during Independence Sale from 4th to 11th July 2022. Buy your dream course at the best price to acquire the #1 skillset of the 21st century! Learn more: opencv.org/courses
@azharrehman15714 ай бұрын
which is current SOTA Image Classification Technique?
@LearnOpenCV4 ай бұрын
Hey, you can check the latest SOTA papers and models on paperswithcode :)
@pablorpalafox2 жыл бұрын
Great overview, thanks Satya!
@xSnipingArtz Жыл бұрын
Thanks for the clear explanation! I am wondering how Model Soups compare to Fast Geometric Ensembling (FGE), Stochastic Weight Averaging (SWA), and Snapshot Ensembles. Could you explain the differences?
@sovitrath4735 Жыл бұрын
Hi. Will research on this topic a bit and get back to you.
@manudasmd7 ай бұрын
Hello can you give a suggestion I am planning to use transfer learning to classify human figure drawings into groups based on certain psychological characteristics into two groups. The drawings i am getting in an A4 paper. My doubts are 1) Should i scan the paper drawings into grey scale format or RGB format 2) Any open source models specifically trained on pencil drawing images available 3) Since my drawings are pencil drawings any specific preprocessing to reduce redundancy of the data?
@kartikpodugu2 жыл бұрын
Sir, what is training protocol? Have never heard before.
@sovitrath4735 Жыл бұрын
Hi. Training protocol in a paper or any practical deep learning training project generally refers to the choices and procedures followed in the training of the model. This can include the hyperparameters like learning rate, batch size, which augmentations to apply and with what probability. In more complex cases this can also include weight initializations and choosing the right optimizer as well.
@golfscienceguru2 жыл бұрын
In a "naming" of objects in images like dog breeds, plant types, mushroom types, and the like, I would imagine that I will score in the low 5% as compared to the around 90+ percent of these super-trained machine models. This is not because my eyes and brain are lacking, but it is because I am lacking in the study of all these hundreds of dog breed, plant types, and mushroom types. Let's say that I will start to develop an edible wild-plant and mushroom computer vision (and with smell) AI model system for my grandkids in preparation for the possible coming of severe global warming, when large parts of the bread-basket lands may be severely impacted by droughts and floods, resulting in widespread famine in the future for my grandkids, who may be forced to forage in the woods for edible wild plants and mushrooms to survive. It is very important for the system to identify poisonous plants and mushroom with an extreme degree of accuracy, especially as pertain to poisonous mushrooms. Some editable mushrooms have lookalikes that are very poisonous. People whose families have been foraging in the wood for generations are experts in detecting poisonous mushrooms, but even them once in a while made fatal mistakes. One type of editable mushroom, and its lookalike highly poisonous version look the same from the top and side views, while it is only the view from the underside shows any difference. Thus, it is necessary to view this mushroom type from multiple angles of views to make the finale judgement. It is kind of like police mug shots with a front and a side view, and with the front view having a ruler measuring the height.
@vaibhavsingh10492 жыл бұрын
This is great, the Model Soup's paper is worth a read.
@PsRafael842 жыл бұрын
Video sugestion: models that can be trained by individuals or small companies. Keep up the good work