fix image sizes which were in tuple

.gitignore

@@ -0,0 +1,14 @@
+
+*.pyc
+
+resnet_640x640/coco_det_finetune_resnet_640x640_checkpoint
+
+resnet_640x640/coco_det_finetune_resnet_640x640_ev_object_detection_validation_p0.4cocoeval.pkl
+
+resnet_640x640/coco_det_finetune_resnet_640x640_ev_object_detection_validation_p0.4_result.json
+
+resnet_640x640/coco_det_finetune_resnet_640x640_config.json
+
+resnet_640x640/coco_det_finetune_resnet_640x640_ckpt-71148.index
+
+resnet_640x640/coco_det_finetune_resnet_640x640_ckpt-71148.data-00000-of-00001

models/ar_model.py

@@ -42,14 +42,14 @@ def __init__(self, config: ml_collections.ConfigDict, **kwargs):
     mlp_ratio = config.dim_mlp // config.dim_att
     if config.resnet_variant == 'c1':
       self.encoder = VisionTransformer(
-          config.image_size[0], config.image_size[1], config.patch_size,
+          config.image_size, config.image_size, config.patch_size,
           config.num_encoder_layers, config.dim_att, mlp_ratio,
           config.num_heads, config.drop_path, config.drop_units,
           config.drop_att, config.pos_encoding, config.use_cls_token,
           name='vit')
     else:
       self.encoder = ResNetTransformer(
-          config.image_size[0], config.image_size[1], config.resnet_variant,
+          config.image_size, config.image_size, config.resnet_variant,
           config.resnet_depth, config.resnet_width_multiplier,
           config.resnet_sk_ratio, config.num_encoder_layers, config.dim_att,
           mlp_ratio, config.num_heads, config.drop_path, config.drop_units,

requirements.txt

@@ -5,4 +5,5 @@ tensorflow-datasets
 tensorflow-addons
 tensorflow-text
 pycocotools
-scikit-image
+scikit-image
+numpy