Adding min_size to classification and video models (#5223)

* Adding min_size as a required field.

* Adding min_size to classification models (quantized and not)

* Adding min_size to video models meta.

* Moving min_size to _COMMON_META

* Fixing extra line
Co-authored-by: Vasilis Vryniotis <datumbox@users.noreply.github.com>

test/test_prototype_models.py

@@ -97,7 +97,7 @@ def test_naming_conventions(model_fn):
 )
 @run_if_test_with_prototype
 def test_schema_meta_validation(model_fn):
-    classification_fields = ["size", "categories", "acc@1", "acc@5"]
+    classification_fields = ["size", "categories", "acc@1", "acc@5", "min_size"]
     defaults = {
         "all": ["task", "architecture", "publication_year", "interpolation", "recipe", "num_params"],
         "models": classification_fields,

torchvision/prototype/models/alexnet.py

@@ -23,6 +23,7 @@ class AlexNet_Weights(WeightsEnum):
             "publication_year": 2012,
             "num_params": 61100840,
             "size": (224, 224),
+            "min_size": (63, 63),
             "categories": _IMAGENET_CATEGORIES,
             "interpolation": InterpolationMode.BILINEAR,
             "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#alexnet-and-vgg",

torchvision/prototype/models/densenet.py

@@ -68,6 +68,7 @@ _COMMON_META = {
     "architecture": "DenseNet",
     "publication_year": 2016,
     "size": (224, 224),
+    "min_size": (29, 29),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
     "recipe": "https://github.com/pytorch/vision/pull/116",

torchvision/prototype/models/efficientnet.py

@@ -66,6 +66,7 @@ _COMMON_META = {
     "task": "image_classification",
     "architecture": "EfficientNet",
     "publication_year": 2019,
+    "min_size": (1, 1),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BICUBIC,
     "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",

torchvision/prototype/models/googlenet.py

@@ -24,6 +24,7 @@ class GoogLeNet_Weights(WeightsEnum):
             "publication_year": 2014,
             "num_params": 6624904,
             "size": (224, 224),
+            "min_size": (15, 15),
             "categories": _IMAGENET_CATEGORIES,
             "interpolation": InterpolationMode.BILINEAR,
             "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#googlenet",

torchvision/prototype/models/inception.py

@@ -23,6 +23,7 @@ class Inception_V3_Weights(WeightsEnum):
             "publication_year": 2015,
             "num_params": 27161264,
             "size": (299, 299),
+            "min_size": (75, 75),
             "categories": _IMAGENET_CATEGORIES,
             "interpolation": InterpolationMode.BILINEAR,
             "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#inception-v3",

torchvision/prototype/models/mnasnet.py

@@ -28,6 +28,7 @@ _COMMON_META = {
     "architecture": "MNASNet",
     "publication_year": 2018,
     "size": (224, 224),
+    "min_size": (1, 1),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
     "recipe": "https://github.com/1e100/mnasnet_trainer",

torchvision/prototype/models/mobilenetv2.py

@@ -23,6 +23,7 @@ class MobileNet_V2_Weights(WeightsEnum):
             "publication_year": 2018,
             "num_params": 3504872,
             "size": (224, 224),
+            "min_size": (1, 1),
             "categories": _IMAGENET_CATEGORIES,
             "interpolation": InterpolationMode.BILINEAR,
             "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#mobilenetv2",

torchvision/prototype/models/mobilenetv3.py

@@ -42,6 +42,7 @@ _COMMON_META = {
     "architecture": "MobileNetV3",
     "publication_year": 2019,
     "size": (224, 224),
+    "min_size": (1, 1),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
 }

torchvision/prototype/models/quantization/googlenet.py

@@ -33,6 +33,7 @@ class GoogLeNet_QuantizedWeights(WeightsEnum):
             "publication_year": 2014,
             "num_params": 6624904,
             "size": (224, 224),
+            "min_size": (15, 15),
             "categories": _IMAGENET_CATEGORIES,
             "interpolation": InterpolationMode.BILINEAR,
             "backend": "fbgemm",

torchvision/prototype/models/quantization/inception.py

@@ -32,6 +32,7 @@ class Inception_V3_QuantizedWeights(WeightsEnum):
             "publication_year": 2015,
             "num_params": 27161264,
             "size": (299, 299),
+            "min_size": (75, 75),
             "categories": _IMAGENET_CATEGORIES,
             "interpolation": InterpolationMode.BILINEAR,
             "backend": "fbgemm",

torchvision/prototype/models/quantization/mobilenetv2.py

@@ -33,6 +33,7 @@ class MobileNet_V2_QuantizedWeights(WeightsEnum):
             "publication_year": 2018,
             "num_params": 3504872,
             "size": (224, 224),
+            "min_size": (1, 1),
             "categories": _IMAGENET_CATEGORIES,
             "interpolation": InterpolationMode.BILINEAR,
             "backend": "qnnpack",

torchvision/prototype/models/quantization/mobilenetv3.py

@@ -66,6 +66,7 @@ class MobileNet_V3_Large_QuantizedWeights(WeightsEnum):
             "publication_year": 2019,
             "num_params": 5483032,
             "size": (224, 224),
+            "min_size": (1, 1),
             "categories": _IMAGENET_CATEGORIES,
             "interpolation": InterpolationMode.BILINEAR,
             "backend": "qnnpack",

torchvision/prototype/models/quantization/resnet.py

@@ -56,6 +56,7 @@ def _resnet(
 _COMMON_META = {
     "task": "image_classification",
     "size": (224, 224),
+    "min_size": (1, 1),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
     "backend": "fbgemm",

torchvision/prototype/models/quantization/shufflenetv2.py

@@ -55,6 +55,7 @@ _COMMON_META = {
     "architecture": "ShuffleNetV2",
     "publication_year": 2018,
     "size": (224, 224),
+    "min_size": (1, 1),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
     "backend": "fbgemm",

torchvision/prototype/models/regnet.py

@@ -50,6 +50,7 @@ _COMMON_META = {
     "architecture": "RegNet",
     "publication_year": 2020,
     "size": (224, 224),
+    "min_size": (1, 1),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
 }

torchvision/prototype/models/resnet.py

@@ -54,6 +54,7 @@ def _resnet(
 _COMMON_META = {
     "task": "image_classification",
     "size": (224, 224),
+    "min_size": (1, 1),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
 }

torchvision/prototype/models/shufflenetv2.py

@@ -45,6 +45,7 @@ _COMMON_META = {
     "architecture": "ShuffleNetV2",
     "publication_year": 2018,
     "size": (224, 224),
+    "min_size": (1, 1),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
     "recipe": "https://github.com/barrh/Shufflenet-v2-Pytorch/tree/v0.1.0",

torchvision/prototype/models/squeezenet.py

@@ -30,6 +30,7 @@ class SqueezeNet1_0_Weights(WeightsEnum):
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_COMMON_META,
+            "min_size": (21, 21),
             "num_params": 1248424,
             "acc@1": 58.092,
             "acc@5": 80.420,
@@ -44,6 +45,7 @@ class SqueezeNet1_1_Weights(WeightsEnum):
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_COMMON_META,
+            "min_size": (17, 17),
             "num_params": 1235496,
             "acc@1": 58.178,
             "acc@5": 80.624,

torchvision/prototype/models/vgg.py

@@ -45,6 +45,7 @@ _COMMON_META = {
     "architecture": "VGG",
     "publication_year": 2014,
     "size": (224, 224),
+    "min_size": (32, 32),
     "categories": _IMAGENET_CATEGORIES,
     "interpolation": InterpolationMode.BILINEAR,
     "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#alexnet-and-vgg",