EndoVisSub2017-RoboticInstrumentSegmentation Banner
Join

Submission Rules

  • When evaluating on a single test set which has an associated training set, exclude those images from your training data.
  • Using additional training data is not allowed, there is however an exception to this when using a publicly available pre-trained CNN. If you design your own CNN and wish to pre-train it, please avoid using any surgical datasets.
  • The submission should be based on the left frame only. 

Submission Format

Binary Segmentation Task

For each image in the test dataset, please submit an identically sized, single channel 8-bit image containing the value 255 at each pixel where you predict a robotic instrument and 0 everywhere else. This means that other surgical objects that may be present (such as drop in probes which are labelled 'Other' in the training set) should be given the value 0. Place these images inside directories which have the same name as the test set and zip the entire contents up in an archive for submission. Name the archive YourName_BinarySegmentation.zip.

Parts Based Segmentation Task

For each image in the test dataset, please submit an identically sized, single channel 8-bit image containing the value 30 for each shaft pixel, 100 for each wrist pixel and 255 for each clasper pixel, these values should be clear from the parts_mappings.json file in the training set. 0 should be predicted for other surgical devices that may be present  (such as drop in probes which are labelled 'Other' in the training set). Place these images inside directories with the same name as the test set and zip the entire contents up in an archive for submission. Name the archive YourName_PartsSegmentation.zip.

Instrument Type Segmentation Task

For each image in the test dataset, please submit an identically sized, single channel 8-bit image containing the unique numerical for each instrument type (which is defined in the type_mapping.json file) at each pixel where you predict a robotic instrument and 0 everywhere else. Objects/instrument types which are not present in this file should be considered background. Place these images inside directories with the same name as the test set and zip the entire contents up in an archive for submission. Name the archive YourName_TypeSegmentation.zip.

Upload

Please upload your submissions here.