Add pre-trained weights for CSPDarkNet

examples/training/classification/imagenet/training_history.json

@@ -1,4 +1,25 @@
 {
+    "cspdarknet": {
+        "v0": {
+            "accelerators": 8,
+            "args": {
+                "batch_size": "64",
+                "epochs": "300",
+                "initial_learning_rate": ".0125",
+                "learning_rate_schedule": "CosineDecayWithWarmup",
+                "warmup_hold_steps_percentage": ".45",
+                "warmup_steps_percentage": ".01"
+            },
+            "contributor": "ianstenbit",
+            "epochs_trained": 299,
+            "script": {
+                "name": "basic_training.py",
+                "version": "dceea23c954e59c5884e98384140e0a8ad5bd320"
+            },
+            "tensorboard_logs": "https://tensorboard.dev/experiment/3eUYhaFMQ3O3fvdg1lDRMw/",
+            "validation_accuracy": "0.7744"
+        }
+    },
     "darknet53": {
         "v0": {
             "accelerators": 8,

keras_cv/models/csp_darknet.py

@@ -30,6 +30,7 @@
 from keras_cv.models.__internal__.darknet_utils import DarknetConvBlockDepthwise
 from keras_cv.models.__internal__.darknet_utils import Focus
 from keras_cv.models.__internal__.darknet_utils import SpatialPyramidPoolingBottleneck
+from keras_cv.models.weights import parse_weights
 
 
 def CSPDarkNet(
@@ -74,8 +75,9 @@ def CSPDarkNet(
             should be used over a regular darknet block. Defaults to False
         classes: optional number of classes to classify images into, only to be
             specified if `include_top` is True.
-        weights: one of `None` (random initialization), or a pretrained weight
-            file path.
+        weights: one of `None` (random initialization), a pretrained weight file
+            path, or a reference to pre-trained weights (e.g. 'imagenet/classification')
+            (see available pre-trained weights in weights.py)
         input_tensor: optional Keras tensor (i.e. output of `layers.Input()`)
             to use as image input for the model.
         input_shape: optional shape tuple, defaults to (None, None, 3).
@@ -95,6 +97,8 @@ def CSPDarkNet(
     Returns:
         A `keras.Model` instance.
     """
+    weights = parse_weights(weights, include_top, "cspdarknet")
+
     if weights and not tf.io.gfile.exists(weights):
         raise ValueError(
             "The `weights` argument should be either `None` or the path to the "

keras_cv/models/weights.py

@@ -39,6 +39,10 @@ def parse_weights(weights, include_top, model_type):
 BASE_PATH = "https://storage.googleapis.com/keras-cv/models"
 
 ALIASES = {
+    "cspdarknet": {
+        "imagenet": "imagenet/classification-v0",
+        "imagenet/classification": "imagenet/classification-v0",
+    },
     "darknet53": {
         "imagenet": "imagenet/classification-v0",
         "imagenet/classification": "imagenet/classification-v0",
@@ -78,6 +82,10 @@ def parse_weights(weights, include_top, model_type):
 }
 
 WEIGHTS_CONFIG = {
+    "cspdarknet": {
+        "imagenet/classification-v0": "8bdc3359222f0d26f77aa42c4e97d67a05a1431fe6c448ceeab9a9c5a34ff804",
+        "imagenet/classification-v0-notop": "9303aabfadffbff8447171fce1e941f96d230d8f3cef30d3f05a9c85097f8f1e",
+    },
     "darknet53": {
         "imagenet/classification-v0": "7bc5589f7f7f7ee3878e61ab9323a71682bfb617eb57f530ca8757c742f00c77",
         "imagenet/classification-v0-notop": "8dcce43163e4b4a63e74330ba1902e520211db72d895b0b090b6bfe103e7a8a5",