Validation errors for ObjectDetectionImageDataConfig class_names

[lcoandrade]

I have a code to perform Object Detection using xView1 Satellite Imagery dataset in YOLO (that is converted to COCO in my notebook) that was working a few months ago. Today, trying to run the same code with the new version I got and error (below) when trying to create a ObjectDetectionImageDataConfig like this:

class_config = ClassConfig(
    names=CLASS_NAMES,
)

#DataConfig – Configure data-related hyper-parameters
data_cfg = ObjectDetectionImageDataConfig(
    class_names=class_config.names,
    class_colors=class_config.colors,
    num_workers=WORKERS, # increase to use multi-processing
    img_channels=IN_CHANNELS,
)

#SolverConfig – Configure the loss, optimizer, and scheduler(s)
solver_cfg = SolverConfig(
    batch_sz=BATCH_SIZE,
    lr=LR,
)

#Model config
# configure the model architecture - by default, RV uses DeepLabV3
# for semantic segmentation
model_config = ObjectDetectionModelConfig(
    backbone=Backbone.resnet50, # The only valid backbones for DeepLabv3 are resnet50 and resnet101
    pretrained=True, # to use imagent pretrained weights (default behaviour)
)

#LearnerConfig – Combine DataConfig, SolverConfig (and optionally, ModelConfig)
learner_cfg = ObjectDetectionLearnerConfig(
    data=data_cfg,
    solver=solver_cfg,
    model=model_config, #using a model config to generate my model automatically
)

This is the error:

---------------------------------------------------------------------------
ValidationError                           Traceback (most recent call last)
Cell In[13], line 6
      1 class_config = ClassConfig(
      2     names=CLASS_NAMES,
      3 )
      5 #DataConfig – Configure data-related hyper-parameters
----> 6 data_cfg = ObjectDetectionImageDataConfig(
      7     class_names=class_config.names,
      8     class_colors=class_config.colors,
      9     num_workers=WORKERS, # increase to use multi-processing
     10     img_channels=IN_CHANNELS,
     11 )
     13 #SolverConfig – Configure the loss, optimizer, and scheduler(s)
     14 solver_cfg = SolverConfig(
     15     batch_sz=BATCH_SIZE,
     16     lr=LR,
     17 )

File /opt/conda/lib/python3.10/site-packages/pydantic/main.py:193, in BaseModel.__init__(self, **data)
    191 # `__tracebackhide__` tells pytest and some other tools to omit this function from tracebacks
    192 __tracebackhide__ = True
--> 193 self.__pydantic_validator__.validate_python(data, self_instance=self)

ValidationError: 2 validation errors for ObjectDetectionImageDataConfig
class_names
  Extra inputs are not permitted [type=extra_forbidden, input_value=['Fixed-wing Aircraft', '...iner', 'Pylon', 'Tower'], input_type=list]
    For further information visit https://errors.pydantic.dev/2.8/v/extra_forbidden
class_colors
  Extra inputs are not permitted [type=extra_forbidden, input_value=[(14, 249, 137), (30, 183... 149, 8), (169, 7, 205)], input_type=list]
    For further information visit https://errors.pydantic.dev/2.8/v/extra_forbidden

How can I solve this?

[lcoandrade]

The update changed to class_config instead of class_names and class_colors.

[AdeelH]

That's right. Sorry for the confusion. Glad you were able to figure it out.

I recommend checking out the migration guides for v0.30 and v0.31.