search

bowenc0221 / panoptic-deeplab

This is Pytorch re-implementation of our CVPR 2020 paper "Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation" (https://arxiv.org/abs/1911.10194)
Apache License 2.0
585 stars 117 forks source link

Feature Request: Add inference-only to train_net.py #82

Closed cmarkle85 closed 3 years ago

cmarkle85 commented 3 years ago

I have been trying to execute one of your models using the detectron2 framework against the cityscapes demoVideo dataset. That dataset is unannotated and I would like to use Panoptic-DeepLab to generate those annotations. I have the dataset registered with the detectron2 framework and I'm using the most of the metadata from the cityscapes-fine dataset. I'm not sure how to generate (and persist) the annotations. I was hoping you could create an inference-only path in train_net.py that would persist the results.

bowenc0221 commented 3 years ago

Detectron2 already has a --eval-only flag. Is this what you are looking for?

cmarkle85 commented 3 years ago

I don't think so. The --eval-only mode returns only the evaluation metrics (it looks like it generates the predicted outputs but saves them to a tempfile that gets erased). I don't have ground truth labels for my dataset. I'm trying to use one of your pretrained models to generate the semantic and instance maps.

bowenc0221 commented 3 years ago

In this case, you will need to write or modify the evaluation metric to save your outputs.