[BUG] apply_model: "Ignoring num_workers parameter; only supported for Torch models" despite using torch models

[daniel-bogdoll]

Describe the problem

While running a set of torch models with apply_model from https://docs.voxel51.com/model_zoo/models.html#torch-models, I noticed that none support the num_workers parameters. This seems odd, since all are torch models. Either the warning message is off or there is something else going on.

Code to reproduce issue

I looked where you check for the usability of num_workers and it seems like a TorchModelMixin type is expected.

Relevant sections seem to be

• fiftyone/fiftyone/core/models.py

  Line 2277 in 1649f2f

  class TorchModelMixin(object):

• fiftyone/fiftyone/core/models.py

  Line 870 in 1649f2f

  use_data_loader = (

However, none of the models seem to have it:

import fiftyone as fo
import fiftyone.zoo as foz
import traceback

dataset = foz.load_zoo_dataset("quickstart", max_samples=10)
model_names = ["detection-transformer-torch", "faster-rcnn-resnet50-fpn-coco-torch", "yolo11l-coco-torch", "yolov5l-coco-torch", "yolov8l-oiv7-torch", "yolov8s-world-torch"]
models = []
for model_name in model_names:
  model = foz.load_zoo_model(model_name)
  models.append(model)

import fiftyone.core.media as fom

# https://github.com/voxel51/fiftyone/blob/1649f2f2e9a98f7499d08921dbc8e9ac050732c2/fiftyone/core/models.py#L2277C1-L2287C9
class TorchModelMixin(object):
    pass

for model in models:
  print(type(model))
  # https://github.com/voxel51/fiftyone/blob/1649f2f2e9a98f7499d08921dbc8e9ac050732c2/fiftyone/core/models.py#L870
  use_data_loader = (isinstance(model, TorchModelMixin) and dataset.media_type == fom.IMAGE)
  print(use_data_loader)

gives

<class 'fiftyone.utils.transformers.FiftyOneTransformerForObjectDetection'>
False
<class 'fiftyone.zoo.models.torch.TorchvisionImageModel'>
False
<class 'fiftyone.utils.ultralytics.FiftyOneYOLODetectionModel'>
False
<class 'fiftyone.utils.torch.TorchImageModel'>
False
<class 'fiftyone.utils.ultralytics.FiftyOneYOLODetectionModel'>
False
<class 'fiftyone.utils.ultralytics.FiftyOneYOLODetectionModel'>
False

So when running

import re
# Does not seem to leverage GPU ?
# Ignores 'num_workers' despite using torch model
for model, model_name in zip(models, model_names):
  label_field = re.sub(r"[\W-]+", "_", "pred_" + model_name)
  dataset.apply_model(model, label_field=label_field, batch_size=8, num_workers=8)

the following warnings appear:

Ignoring `num_workers` parameter; only supported for Torch models
WARNING:fiftyone.core.models:Ignoring `num_workers` parameter; only supported for Torch models

So either by "Torch models" you do not mean the ones you provide in the Zoo, or something is off.

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 22.04): Google Colab T4 Instance
• Python version (python --version): Python 3.10.12
• FiftyOne version (fiftyone --version): FiftyOne v1.0.2, Voxel51, Inc.
• FiftyOne installed from (pip or source): pip

Willingness to contribute

The FiftyOne Community encourages bug fix contributions. Would you or another
member of your organization be willing to contribute a fix for this bug to the
FiftyOne codebase?

• Yes. I can contribute a fix for this bug independently
• Yes. I would be willing to contribute a fix for this bug with guidance
  from the FiftyOne community
• No. I cannot contribute a bug fix at this time