Change Log


  • Fixed python API bug in Object Detection model load function that caused .h5 file to be modified when .h5 file belonging to different model was accidentally loaded
  • Added support for images with “.JPG” file extension


  • Fixed python API bug in GradCAM class where object initialization caused an overflow
  • Fixed python API bug in GradCAM where class index of 0 on GPU caused an indexing error


  • Fixed python API bug in EvaluatorSegmentation class so that num_class argument is properly accepted by function


  • Modified weight initialization method in SSD conv2d layers
  • Modified max_pool2d layers in SSD to use ceiling mode for certain layers
  • Moved setting of schedule in optimizer to before gradient update step in for GUI training to match API fit method
  • Fixed segmentation prediction result RGB colors


  • Modified UNet architecture in python API to match official paper and added mirror padding option
  • Fixed Docker file pip version issue by adding upgrade option to pip install
  • Added constraints to Yolov1, Yolov2 anchor and cell input parameters on GUI


  • Added Classifier and Segmenter inference modules for downloading classification and segmentation models deployed on GUI and running predictions on local images
  • Added DenseNet-121, DenseNet-169 and DenseNet-201 classification models to GUI
  • Revised DenseNet BatchNormalize layer parameters to match official paper
  • Added DenseNet-C10 model for small images sizes (such as 32x32) to python API
  • Made pretrained weights for all DenseNet models available for download when instantiating models
  • Reversed order of RGB channels in SSD and VGG pre-processing to match channels defined in pretrained weights
  • Changed node.js installation version from ‘stable’ to ‘v10.17.0’ in Docker file to avoide incompatibility between node.js and node-sass versions
  • Fixed GUI bug where segmentation prediction result colors as shown on image did not much colors for the corresponding class in the tag list
  • Rounded classification prediction score values to 4 decimal places for model.predict() function
  • Rounded classification prediction score values to 4 decimal places inside GUI Classification Prediction page CSV output
  • Added explanation about GUI Learning Curve plot to documentation


  • Fixed bug where model ID information was displayed as ‘undefined’ when stopping model training on GUI
  • Fixed bug in GUI Prediction Result display where ‘Image’ toggle status on Segmentation Page affected whether or not images were displayed on the Classification and Detection pages
  • Modified error message displayed on GUI when training diverges and model parameters cause a numerical overflow to be more descriptive and suggest alternative actions to take
  • Fixed incorrect batch normalization layer momentum value for Yolo v1, Yolo v2, ResNet and ResNeXt models
  • Fixed bug in Yolo v2 Python API loss function that caused error when image height and width were not equal to each other


  • Fixed bug in Grad-CAM API that caused error when calculating forward propagation in model
  • Fixed bug in forward propagation method definition that allowed model to calculate output even when number of classes were undefined


  • Added prediction scores column to prediction.csv output on “Predict” GUI page for Classification models
  • Changed Model Distribution plot to only show blinking point for model that is currently training
  • Changed Model Distribution plot to remove points for model in CREATED or RESERVED states
  • Fixed bug in classmap loading function for SSD object detection algorithm
  • Fixed bug in Deeplabv3+ segmentation algorithm classmap definition
  • Fixed bug in model state where model would not enter RESERVED state


  • Added Deeplabv3+ segmentation model to GUI
  • Revised GUI component details


  • Made GUI text selectable
  • Fixed bugs


  • Added Deeplabv3+ segmentation model to API
  • Refactored GUI components
  • Refactored backend server
  • Refactored CNN model architecture code
  • Modified Yolov2 loss function
  • Modified Yolov1 and Yolov2 pretrained weights
  • Fixed bugs


  • Fixed bugs
  • Revised documentation


  • Added Grad-CAM visualization tool
  • Added pytests
  • Revised format
  • Added URLs to download past wheel packages to
  • Fixed bugs


  • Modify img loader to accept binary image


  • Add new augmentation methods
  • Add a function for downloading the prediction result as csv
  • Modify image data preprocess pipeline
  • Update Node.js and Python dependencies
  • Fix UI Bugs


  • Update dependencies


  • Fix UI Bugs


  • Add warning to dataset create modal(GUI) when illegal train valid ratio is inputted
  • Add error handler to renom_img.api.inference.detector.Detector


  • Fixed UI bugs
  • Updated webpack @ 3 . x => webpack @ 4 . x
  • Modified eslint settings
  • Modified Darknet architecture (Added BatchNormalization and removed bias term from convolution layer)
  • Modified Yolov1 loss function