"No one is harder on a talented person than the person themselves" - Linda Wilkinson ; "Trust your guts and don't follow the herd" ; "Validate direction not destination" ;

February 13, 2022

Fundamentals Revisited

Question #1 CNN - Why Convolution 2D?

2D convolution refers *not* to the dimension of the convolution kernel but to the dimension of the output. The output’s dimension is 2D, single channel. A single 2D convolution pass over a 3D image uses a 3D convolution kernel to obtain the 2D output

Conv2D: Input-->One filter --> 2D output

Conv3D: Input--> One filter --> 3D output

Ref - Link

Question #2 - Visualize Layers


The output across multiple layers. Link

Question #3 - Different types of Data Augmentation

Ref - Link

Keep Exploring!!!

February 10, 2022

Paper - Class-Balanced Loss Based on Effective Number of Samples

  • In general, there are two strategies: re-sampling and cost-sensitive re-weighting
  • In re-sampling, the number of examples is directly adjusted by over-sampling
  • In cost-sensitive re-weighting, we influence the loss function by assigning relatively higher costs to examples from minor classes softmax cross-entropy, sigmoid cross-entropy and focal loss.

Code - Link

Keep Exploring!!!

Fashion Segmentation Paper Read

Paper - U2-Net: Going Deeper with Nested U-Structure for Salient Object Detection

Key Notes

  • Capture more contextual information from different scales thanks
  • Code link https://github.com/xuebinqin/U-2-Net
  • Segmenting the most visually attractive objects in an image
  • Deep features extracted by existing backbones, such as Alexnet [17], VGG [35], ResNet [12], ResNeXt [44], DenseNet [15]
  • Convolution with stride of two followed by a maxpooling with stride of two are utilized to reduce the size of the feature maps to one fourth
  • Go deeper while maintaining high resolution feature maps
  • ReSidual U-block (RSU), which is able to extract intra-stage multi-scale features 
  • Multi-scale feature extraction - A 3 × 3 filter is good for extracting local features at each layer
  • Convolution + Feature Extraction + Downsample + Upsample

  • multi-scale feature extraction target at designing new modules for extracting both local and global information from features obtained by backbone networks.

RSU mainly consists of three component (ReSidual Ublock, RSU)

  • an input convolution layer, which transforms the input feature map
  • a U-Net like symmetric encoder-decoder structure which takes the intermediate feature map as input and learns to extract and encode the multi-scale contextual information
  • a residual connection which fuses local features and the multi-scale features

Dataset - Link

  • Labelled Images Samples
  • Ground Truth / Training Images
  • After 600k iterations (with a batch size of 12), the training loss converges and the whole training process takes about 120 hours Sample data


Creating this is also the key

Usecases

  • Remove background
  • Create portrait view

Paper #2 - BASNet: Boundary-Aware Salient Object Detection

Code - Link

Background removal tool - Link

Notes

  • Architecture is composed of a densely supervised Encoder-Decoder network and a residual refinement module
  • Hybrid loss - Binary Cross Entropy (BCE), Structural SIMilarity (SSIM) and Intersectionover-Union (IoU) losses.
  • Code Link
  • It assembles a UNet-like [57] deeply supervised [31, 67] Encoder-Decoder network with a novel residual refinement module

Paper #3 - BiSeNet: Bilateral Segmentation Network for Real-time Semantic Segmentation

Notes

  • Spatial Path with a small stride to preserve the spatial information and generate high-resolution features
  • Context Path with a fast downsampling strategy is employed to obtain sufficient receptive field
  • Spatial Path (SP) and Context Path (CP)


Paper #4 - BiSeNet V2: Bilateral Network with Guided Aggregation for Real-time Semantic Segmentation

Notes

This architecture involves: (i) a Detail Branch, with wide channels and shallow layers to capture low-level details and generate high-resolution feature representation; (ii) a Semantic Branch, with narrow channels and deep layers to obtain high-level semantic context

Keep Exploring!!!

February 09, 2022

Human Body and Emotions

 Paper - Link

Different emotions and activation observed in human body

Emotions and their behavior/combinations

Keep Exploring!!!

January 31, 2022

Colab Experiment Tips

 


Keep Exploring

January 30, 2022

Metaverse

Every new technology sometimes I feel I need to understand out of FOMO - Fear of missing out. Metaverse - Meta represents beyond, Verse represents the universe, beyond the universe. Ideally, different components of metaverse existed earlier. We had AR, VR, AI, Robotics, Computer Vision, and adoption has been successful/ongoing. A system that would encompass everything and layout a virtual platform is a metaverse. 

Today you collaborate with the Mural board, Teams, Whiteboard. Oculus, Hololens, Google glass many of these projects tried to project some form of information into the real world. Something similar if it could encompass and provide Engagement, experience, and emotions is all what metaverse to go about.

Microsoft, Facebook, Google have all been trying to get a break on this. We know what happened to google glass, What happed to Nokia. Some of the projects end up as kitchen sink. They never make it to production. Although there are benefits of these components the First-mover cost, experimentation, getting the tech breakthrough may involve a lot more costs. 

Essentially when it becomes affordable it may end up as another commodity device, a play store kind of ecosystem where apps get published.

5G may power provide the required network speed but the rest of AR, VR, Robotics, Vision, Blockchain may still need some iterations to get some use case beyond what we have individually for AR, VR, Vision, Robotics etc.

Keep Thinking!!!

January 28, 2022

Interesting JD - DataScience Roles - Evaluation Software Engineer

I have come across several d#atascience #jobs, This one Evaluation Software Engineer is very interesting

The JD States below things

  • Organize neural network challenge scenarios and route them to the appropriate evaluation suites
  • Collaborate with engineers and program managers to identify which neural network challenge cases are the highest priority to improve
  • Investigate if the challenge persists in newer versions of models

My version of understanding

For a vision model for a failed use case - pedestrian not detected, vehicle not detected they triage / prioritize / address

  • Why a use case fails, How to enrich the dataset
  • Do the key regions are activated when we interpret feature activation across layers
  • Prioritize / Add data / customize network if needed / train / validate it is fixed

This reflects how much every scenario is validated, prioritized, and ensured models reflect the real-world scenarios. Most of the time we see ML, DL jobs but not this level of details and clarity.

The JD link

This is the difference between prototype vs production vs updates and how forward-looking they are in the future to handle all scenarios :), Behind all #autopilot models there would be tons of #scenarios and multiple Evaluation Software Engineers and automated suites validating it.

I have never seen a similar type of JD anywhere except Tesla :)

Keep Exploring!!

January 27, 2022

Dart based forecasting

Few experiments Transformer based / Exponential based

Time Series Made Easy in Python


More reads

Key concepts

  • Past Covariates denote time series whose past values are known at prediction time.
  • Future Covariates These can for instance represent known future holidays, or weather forecasts.
Multi-step Time Series Forecasting with Machine Learning
Multivariate-Time-series-Analysis-using-LSTM-ARIMA
Multivariate-Time-Series-Forecasting-with-LSTMs-in-Keras-for-CORN-SWEET-Terminal-Market-Price

Keep Exploring!!!

surgeahead - Airavana, BambooBox, Blend , #hellomida - Perspectives with ML Lens

What are potential ML use cases powering this use cases, Lets discuss and explore

Keep Exploring!!

January 26, 2022

Career, Growth, Trust

Why do we switch careers?

  • Better Salary
  • Better Job profiles
  • Relocation
  • Brand

What drives you?

  • Unsure of growth in the current place
  • Not getting recognition

What is unsure of growth?

  • The trust deficit between you and your manager
  • I may not grow here or I don't sense my growth here

How you may grow?

  • When you get your right projects
  • When your work is appreciated
  • It has both internal and external factors
  • Your skills, projects, Contributions

Does it end with a Job Change?

  • New job, the new team may be the same story
  • Match of skills
  • Culture, Able to question and ask with clarity
  • Contribute, question, collaborate 

In long term, you are your own competition. Titles - Jobs - Salary is point in time but what you learn, build, make your own uniqueness, strength is the key.

Good read - link

  • Transparently sharing it without request
  • Make it clear that it's just opinions and not decisions

Keep Thinking!!!