Merge pull request #577 from jhwisdom/hut101-19-jhwisdom

Hut101 19 jhwisdom

Author: rwood-97

CHANGELOG.md

@@ -22,6 +22,7 @@ _Add new changes here_
 
 ### Added
 
+- Added Hugging Face model support to ClassifierContainer
 - Added `piffle` package as dependency ([#575](https://github.com/maps-as-data/MapReader/pull/575))
 
 ## [v1.8.1](https://github.com/Living-with-machines/MapReader/releases/tag/v1.8.1) (2025-08-11)

docs/source/using-mapreader/step-by-step-guide/4-classify/train.rst

@@ -344,25 +344,24 @@ There are a number of options for the ``model`` argument:
 
             If you use this option, your optimizer, scheduler and loss function will be loaded from last time.
 
-    **4.  To load a** `hugging face model <https://huggingface.co/models>`__\ **, choose your model, follow the "Use in Transformers" or "Use in timm" instructions to load it and then pass this as the ``model`` argument.**
+    **4.  To load a** `hugging face model <https://huggingface.co/models>`__\ **, pass the model's repository ID as a string and set ``huggingface=True``.**
 
-        e.g. `This model <https://huggingface.co/davanstrien/autotrain-mapreader-5000-40830105612>`__ is based on our `*gold standard* dataset <https://huggingface.co/datasets/Livingwithmachines/MapReader_Data_SIGSPATIAL_2022>`__.
-        It can be loaded using the `transformers <https://github.com/huggingface/transformers>`__ library:
+        MapReader will automatically download the model and its corresponding image processor from the Hugging Face Hub using the `transformers <https://github.com/huggingface/transformers>`__ library.
 
+        e.g. `This model <https://huggingface.co/davanstrien/autotrain-mapreader-5000-40830105612>`__ is based on our `*gold standard* dataset <https://huggingface.co/datasets/Livingwithmachines/MapReader_Data_SIGSPATIAL_2022>`__.
+        It can be loaded directly like this: 
+        
         .. code-block:: python
 
             #EXAMPLE
             import torch
-            from transformers import AutoFeatureExtractor, AutoModelForImageClassification
-
             from mapreader import ClassifierContainer
 
-            extractor = AutoFeatureExtractor.from_pretrained("davanstrien/autotrain-mapreader-5000-40830105612")
-            my_model = AutoModelForImageClassification.from_pretrained("davanstrien/autotrain-mapreader-5000-40830105612")
+            my_model = "davanstrien/autotrain-mapreader-5000-40830105612"
 
             device = 'cuda' if torch.cuda.is_available() else 'mps' if torch.backends.mps.is_available() else 'cpu'
 
-            my_classifier = ClassifierContainer(my_model, annotated_images.labels_map, dataloaders, device=device)
+            my_classifier = ClassifierContainer(my_model, annotated_images.labels_map, dataloaders, device=device, huggingface=True)
 
         .. note:: You will need to install the `transformers <https://github.com/huggingface/transformers>`__ library to do this (``pip install transformers``).
 

mapreader/classify/classifier.py

@@ -108,6 +108,7 @@ def __init__(
         is_inception: bool = False,
         load_path: str | None = None,
         force_device: bool = False,
+        huggingface: bool = False,
         **kwargs,
     ):
         # set up device
@@ -149,7 +150,31 @@ def __init__(
                 self.input_size = input_size
                 self.is_inception = is_inception
             elif isinstance(model, str):
-                self._initialize_model(model, **kwargs)
+                if huggingface:
+                    try:
+                        from transformers import AutoModelForImageClassification, AutoImageProcessor
+                    except ImportError:
+                        raise ImportError(
+                            "Hugging Face models require the 'transformers' library: 'pip install transformers'."
+                        )
+                    print(f"[INFO] Initializing Hugging Face model: {model}")
+                    num_labels = len(self.labels_map)
+                    self.model = AutoModelForImageClassification.from_pretrained(
+                        model,
+                        num_labels=num_labels,
+                        ignore_mismatched_sizes=True
+                    ).to(self.device)
+                    hf_processor = AutoImageProcessor.from_pretrained(model)
+                    size = getattr(hf_processor, "size", {})
+                    if "height" in size and "width" in size:
+                        size = (size["height"], size["width"])
+                    elif "shortest_edge" in size:
+                        size = (size["shortest_edge"], size["shortest_edge"])
+                    else:
+                        size = input_size
+                    self.is_inception = False
+                else:
+                    self._initialize_model(model, **kwargs)
 
             self.optimizer = None
             self.scheduler = None