Clean up purely informational fields from Weight Meta-data (#5852)

* Removing `task`, `architecture` and `quantization`

* Fix mypy

* Remove size field

* Remove unused import.

* Fix mypy

* Remove size from schema list.

* update todo

* Simplify with assert

* Adding min_size to all models.

* Update RAFT min size to 128

Author: datumbox

test/test_extended_models.py

@@ -79,12 +79,13 @@ def test_naming_conventions(model_fn):
 )
 @run_if_test_with_extended
 def test_schema_meta_validation(model_fn):
-    classification_fields = ["size", "categories", "acc@1", "acc@5", "min_size"]
+    # TODO: add list of permitted fields
+    classification_fields = ["categories", "acc@1", "acc@5"]
     defaults = {
-        "all": ["task", "architecture", "recipe", "num_params"],
+        "all": ["recipe", "num_params", "min_size"],
         "models": classification_fields,
         "detection": ["categories", "map"],
-        "quantization": classification_fields + ["backend", "quantization", "unquantized"],
+        "quantization": classification_fields + ["backend", "unquantized"],
         "segmentation": ["categories", "mIoU", "acc"],
         "video": classification_fields,
         "optical_flow": [],

torchvision/models/alexnet.py

@@ -57,10 +57,7 @@ class AlexNet_Weights(WeightsEnum):
         url="https://download.pytorch.org/models/alexnet-owt-7be5be79.pth",
         transforms=partial(ImageClassification, crop_size=224),
         meta={
-            "task": "image_classification",
-            "architecture": "AlexNet",
             "num_params": 61100840,
-            "size": (224, 224),
             "min_size": (63, 63),
             "categories": _IMAGENET_CATEGORIES,
             "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#alexnet-and-vgg",

torchvision/models/convnext.py

@@ -204,9 +204,6 @@ def _convnext(
 
 
 _COMMON_META = {
-    "task": "image_classification",
-    "architecture": "ConvNeXt",
-    "size": (224, 224),
     "min_size": (32, 32),
     "categories": _IMAGENET_CATEGORIES,
     "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#convnext",

torchvision/models/densenet.py

@@ -266,9 +266,6 @@ def _densenet(
 
 
 _COMMON_META = {
-    "task": "image_classification",
-    "architecture": "DenseNet",
-    "size": (224, 224),
     "min_size": (29, 29),
     "categories": _IMAGENET_CATEGORIES,
     "recipe": "https://github.com/pytorch/vision/pull/116",

torchvision/models/detection/faster_rcnn.py

@@ -370,9 +370,8 @@ def forward(self, x):
 
 
 _COMMON_META = {
-    "task": "image_object_detection",
-    "architecture": "FasterRCNN",
     "categories": _COCO_CATEGORIES,
+    "min_size": (1, 1),
 }
 
 

torchvision/models/detection/fcos.py

@@ -651,10 +651,9 @@ class FCOS_ResNet50_FPN_Weights(WeightsEnum):
         url="https://download.pytorch.org/models/fcos_resnet50_fpn_coco-99b0c9b7.pth",
         transforms=ObjectDetection,
         meta={
-            "task": "image_object_detection",
-            "architecture": "FCOS",
             "num_params": 32269600,
             "categories": _COCO_CATEGORIES,
+            "min_size": (1, 1),
             "recipe": "https://github.com/pytorch/vision/tree/main/references/detection#fcos-resnet-50-fpn",
             "map": 39.2,
         },

torchvision/models/detection/keypoint_rcnn.py

@@ -308,10 +308,9 @@ def forward(self, x):
 
 
 _COMMON_META = {
-    "task": "image_object_detection",
-    "architecture": "KeypointRCNN",
     "categories": _COCO_PERSON_CATEGORIES,
     "keypoint_names": _COCO_PERSON_KEYPOINT_NAMES,
+    "min_size": (1, 1),
 }
 
 

torchvision/models/detection/mask_rcnn.py

@@ -351,9 +351,8 @@ def __init__(self, in_channels, dim_reduced, num_classes):
 
 
 _COMMON_META = {
-    "task": "image_object_detection",
-    "architecture": "MaskRCNN",
     "categories": _COCO_CATEGORIES,
+    "min_size": (1, 1),
 }
 
 

torchvision/models/detection/retinanet.py

@@ -674,9 +674,8 @@ def forward(self, images, targets=None):
 
 
 _COMMON_META = {
-    "task": "image_object_detection",
-    "architecture": "RetinaNet",
     "categories": _COCO_CATEGORIES,
+    "min_size": (1, 1),
 }
 
 

torchvision/models/detection/ssd.py

@@ -30,11 +30,9 @@ class SSD300_VGG16_Weights(WeightsEnum):
         url="https://download.pytorch.org/models/ssd300_vgg16_coco-b556d3b4.pth",
         transforms=ObjectDetection,
         meta={
-            "task": "image_object_detection",
-            "architecture": "SSD",
             "num_params": 35641826,
-            "size": (300, 300),
             "categories": _COCO_CATEGORIES,
+            "min_size": (1, 1),
             "recipe": "https://github.com/pytorch/vision/tree/main/references/detection#ssd300-vgg16",
             "map": 25.1,
         },