testing for segment_weeds.py

[mkutu]

1. save results by species for a more thorough/easier way to identify segmentation cleaning differences
2. play around/ look into pixelated borders that's occurring for some "broad" categories
   • create rules to define area_threshold in remove_small_holes for example:

     if clean_mask_area > 1000:
     area_threshold = 100

3. clean up the script
   • organize saving and cleaning in separate methods
   • place remove_gray_hsv_color in clean_mask
   • add comments in large blocks of code
4. place "broad" and "sparse" class_ids in config file, not hard coded in the script with their common name in comments
5. add "category" in species metadata. for examples:

   {
   "detection_results": {
       "image_id": "DSC09709",
       "bbox": {
           "x_min": 457,
           "y_min": 25,
           "x_max": 1295,
           "y_max": 1066
       },
       "class_id": 41
   },
   "category": {
       "SEOB4": {
           "class_id": 41,
           "USDA_symbol": "SEOB4",
           "EPPO": "CASOB",
           "group": "dicot",
           "class": "Magnoliopsida",
           "subclass": "Rosidae",
           "order": "Fabales",
           "family": "Fabaceae",
           "genus": "Senna",
           "species": "obtusifolia",
           "common_name": "Sicklepod",
           "authority": "(Linnaeus) Irwin and Barneby",
           ...
       }
   },....

6. explore moving remove_gray_hsv before remove_holes
7. explore remove_objects or combining with remove_holes
8. check the 2 list of sparse vs broad to make sure you've got all the weeds

[navjot-nangia]

Remaining:

1. play around/ look into pixelated borders that's occurring for some "broad" categories

2. create rules to define area_threshold in remove_small_holes for example: if clean_mask_area > 1000: area_threshold = 100

3. clean up the script

   • organize saving and cleaning in separate methods
   • place remove_gray_hsv_color in clean_mask
   • add comments in large blocks of code

4. explore moving remove_gray_hsv before remove_holes

5. explore remove_objects or combining with remove_holes

[navjot-nangia]

Remaining:

1. play around/ look into pixelated borders that's occurring for some "broad" categories
2. create rules to define area_threshold in remove_small_holes for example: if clean_mask_area > 1000: area_threshold = 100
3. clean up the script organize saving and cleaning in separate methods add comments in large blocks of code
4. explore remove_objects or combining with remove_holes