---

name: core-apis description: "Routes Mask_RCNN core API, Config, Dataset, MaskRCNN, utility, visualization, and TensorFlow/Keras compatibility tasks." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Core APIs

Use this sub-skill when a task asks how to use or debug the installed mrcnn package itself: configuration classes, MaskRCNN construction, utility functions, visualization signatures, graph-building constraints, or legacy TensorFlow/Keras compatibility.

Read first by need

• Read api-reference.md for verified public signatures and object relationships.
• Read compatibility.md before choosing TensorFlow/Keras versions or interpreting modern Keras errors.
• Read troubleshooting.md when imports, graph construction, image dimensions, or API calls fail.
• Run scripts/inspect_mask_rcnn_api.py to inspect a user environment without requiring the original checkout.

Core API workflow

1. Verify the runtime stack and import package:

   python sub-skills/core-apis/scripts/inspect_mask_rcnn_api.py --show-signatures
   

2. Define a Config subclass. Set at least NAME, NUM_CLASSES, image size/resize mode, GPU_COUNT, and IMAGES_PER_GPU. Dimensions that flow into the FPN should be divisible by 64.

3. Construct the model with a mode-specific object:

   from mrcnn.config import Config
   from mrcnn import model as modellib
   
   class MyConfig(Config):
       NAME = "my_project"
       NUM_CLASSES = 1 + 1
       GPU_COUNT = 1
       IMAGES_PER_GPU = 1
       IMAGE_MIN_DIM = 512
       IMAGE_MAX_DIM = 512
   
   config = MyConfig()
   model = modellib.MaskRCNN(mode="inference", config=config, model_dir="logs")
   

4. For training, route to training. For dataset subclasses and masks, route to data-preparation. For detect() results and visualization, route to inference-evaluation.

Decision points

• Using modern Keras? Treat failures as compatibility/porting unless a full graph build and target workflow already pass. The legacy stack is the safer operating default.
• CPU versus GPU? CPU can validate imports, shapes, utilities, and tiny graphs. Real training and multi-GPU behavior need backend-specific runtime verification.
• Weights mismatch? Core API can explain load_weights; training owns the transfer-learning recipes and layer exclusions.
• Original sample path requested? Use bundled references/scripts instead. This repo skill is meant to operate without reopening the source checkout.