Deep Learning Final Project 4995
Based largely on:
- Learning to Segment Every Thing
Output a image with segmap/bbox based on input query object(s)
python2 infer_query_2.py \ --input_query glass \ --output-dir kid_bat/query_seg\ --image-ext jpg \ --thresh 0.5 --use-vg3k \ kid_bat/inputif no object is detected in the image, orginal input image will be output.
Description of code files:
- infer_query_2.py - infer the segmentation map and bounding box of input image based on input query object
- infer_simple.py - infer the segmentation mape and bounding boxes of all objects of input image.
- dummy_datasets.py - include all classes of objects in VG dataset.
- build_dict.py - helper function to build a dictionary for indexing all classes in VG3K classes of VG dataset.
- class_filter.py - helper function to set the bounding box of a object detected not present in input query object list to empy.
- vis.py - visualize the image with segmentation map and bounding box.
Description of data folder:
- VG_data_p2 - original images
- p2_all_seg - output for segmentation for all objects in VG3K classes
- p2_query_seg - output for segmentation for a subset of objects from input query
Description of dataset:
- Visual genome dataset - https://visualgenome.org/
- Download - https://visualgenome.org/api/v0/api_home.html
Dependencies:
- caffe2 - https://github.com/ronghanghu/seg_every_thing/blob/master/INSTALL.md
- detecteron - https://github.com/facebookresearch/Detectron
- numpy
- python 2.7
- OpenCV
- COCO API - https://github.com/cocodataset/cocoapi