predict() support for segmentation models and more (#1708)

* squashed commits from segmentation predict branch * Set ignore_errors=True to trainer test and declare why * version bumped (#1374) Co-authored-by: Eugene Khvedchenya <ekhvedchenya@gmail.com> * segmentation predict: remove predict from ppliteseg models (PPLiteSegBase) as it's implement in SegmentationModule * Added missing transforms in validation * Add missing default preprocessing params for rest of the models * Add STDC predict * Update the predict & transforms & deprecate SegToTensor * Added lru_cache(1) to _get_pipeline method * Use only URL in the example script * Rename segmentation processing classes to have "Segmentation" prefix * Remove test for predict using segformer b0 * Segmentation transforms to use nd.ndarray whenever possible * Segmentation transforms to use nd.ndarray whenever possible * Added tests * Added tests * Fix typos * Merge master --------- Co-authored-by: dafna-r <dalia.rifold@gmail.com> Co-authored-by: dafna-r <47064033+dafna-r@users.noreply.github.com> Co-authored-by: Louis-Dupont <35190946+Louis-Dupont@users.noreply.github.com> Co-authored-by: Shay Aharon <80472096+shaydeci@users.noreply.github.com> Co-authored-by: Louis Dupont <louis-dupont@live.fr>
Deci-AI · Dec 20, 2023 · c4bbce7 · c4bbce7
1 parent 445db0f
commit c4bbce7
Show file tree

Hide file tree

Showing 40 changed files with 2,044 additions and 1,698 deletions.
diff --git a/notebooks/transfer_learning_semantic_segmentation.ipynb b/notebooks/transfer_learning_semantic_segmentation.ipynb
diff --git a/src/super_gradients/common/object_names.py b/src/super_gradients/common/object_names.py
@@ -42,7 +42,9 @@ class Transforms:
 
     # From SG
     SegNormalize = "SegNormalize"
+    SegStandardize = "SegStandardize"
     SegToTensor = "SegToTensor"
+    SegConvertToTensor = "SegConvertToTensor"
     SegRandomFlip = "SegRandomFlip"
     SegResize = "SegResize"
     SegRescale = "SegRescale"
@@ -53,6 +55,7 @@ class Transforms:
     SegPadShortToCropSize = "SegPadShortToCropSize"
     SegPadToDivisible = "SegPadToDivisible"
     SegColorJitter = "SegColorJitter"
+    # Detection transforms
     DetectionStandardize = "DetectionStandardize"
     DetectionMosaic = "DetectionMosaic"
     DetectionRandomAffine = "DetectionRandomAffine"
@@ -442,3 +445,8 @@ class Processings:
     ComposeProcessing = "ComposeProcessing"
     CenterCrop = "CenterCrop"
     Resize = "Resize"
+    SegmentationResizeWithPadding = "SegmentationResizeWithPadding"
+    SegmentationRescale = "SegmentationRescale"
+    SegmentationResize = "SegmentationResize"
+    SegmentationPadShortToCropSize = "SegmentationPadShortToCropSize"
+    SegmentationPadToDivisible = "SegmentationPadToDivisible"
diff --git a/src/super_gradients/examples/predict/segmentation_predict.py b/src/super_gradients/examples/predict/segmentation_predict.py
@@ -0,0 +1,13 @@
+from super_gradients.common.object_names import Models
+from super_gradients.training import models
+
+
+model = models.get(model_name=Models.PP_LITE_T_SEG75, pretrained_weights="cityscapes")
+
+IMAGES = [
+    "https://deci-pretrained-models.s3.amazonaws.com/sample_images/beatles-abbeyroad.jpg",
+]
+
+predictions = model.predict(IMAGES)
+predictions.show()
+predictions.save(output_folder="")  # Save in working directory
diff --git a/src/super_gradients/recipes/cityscapes_al_ddrnet.yaml b/src/super_gradients/recipes/cityscapes_al_ddrnet.yaml
@@ -57,12 +57,16 @@ dataset_params:
           crop_size: [ 1024, 1024 ]
           mode: random
 
-      - SegToTensor
+      - SegStandardize:
+          max_value: 255
 
       - SegNormalize:
           mean: [0.485, 0.456, 0.406]
           std: [0.229, 0.224, 0.225]
 
+      - SegConvertToTensor
+
+
 
 training_hyperparams:
   max_epochs: 200

diff --git a/src/super_gradients/recipes/dataset_params/cityscapes_al_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/cityscapes_al_dataset_params.yaml
@@ -19,22 +19,32 @@ train_dataset_params:
   cache_labels: False
   cache_images: False
   transforms:
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
+
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 val_dataset_params:
   root_dir: /data/cityscapes
   list_file: lists/val.lst
   labels_csv_path: lists/labels.csv
   cache_labels: False
   cache_images: False
   transforms:
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
+
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
+
+    - SegConvertToTensor
+
 train_dataloader_params:
   dataset: CityscapesConcatDataset
   shuffle: True

diff --git a/src/super_gradients/recipes/dataset_params/cityscapes_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/cityscapes_dataset_params.yaml
@@ -5,10 +5,13 @@ train_dataset_params:
   cache_labels: False
   cache_images: False
   transforms:
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
+    - SegConvertToTensor
+
 
 val_dataset_params:
   root_dir: /data/cityscapes
@@ -17,10 +20,12 @@ val_dataset_params:
   cache_labels: False
   cache_images: False
   transforms:
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
+    - SegConvertToTensor
 
 train_dataloader_params:
   shuffle: True

diff --git a/src/super_gradients/recipes/dataset_params/cityscapes_ddrnet_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/cityscapes_ddrnet_dataset_params.yaml
@@ -23,12 +23,16 @@ train_dataset_params:
         crop_size: [ 1024, 1024 ]
         mode: random
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 train_dataloader_params:
   batch_size: 6
 

diff --git a/src/super_gradients/recipes/dataset_params/cityscapes_ppliteseg_seg75_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/cityscapes_ppliteseg_seg75_dataset_params.yaml
@@ -24,17 +24,31 @@ train_dataset_params:
         crop_size: [ 768, 768 ]
         mode: random
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 val_dataset_params:
   transforms:
     - SegRescale:
         scale_factor: 0.75
 
+    - SegStandardize:
+        max_value: 255
+
+    - SegNormalize:
+        mean: [0.485, 0.456, 0.406]
+        std: [0.229, 0.224, 0.225]
+
+    - SegConvertToTensor
+
+
 train_dataloader_params:
   batch_size: 8
 

diff --git a/src/super_gradients/recipes/dataset_params/cityscapes_regseg48_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/cityscapes_regseg48_dataset_params.yaml
@@ -31,12 +31,16 @@ train_dataset_params:
         crop_size: 1024
         mode: random
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 train_dataloader_params:
   batch_size: 4
   num_workers: 0

diff --git a/src/super_gradients/recipes/dataset_params/cityscapes_segformer_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/cityscapes_segformer_dataset_params.yaml
@@ -23,12 +23,16 @@ train_dataset_params:
         crop_size: [ 1024, 1024 ]
         mode: random
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 val_dataset_params:
   transforms:
@@ -39,12 +43,16 @@ val_dataset_params:
         crop_size: [ 1024, 1024 ]
         fill_mask: 19
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 train_dataloader_params:
   batch_size: 2

diff --git a/src/super_gradients/recipes/dataset_params/cityscapes_stdc_seg50_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/cityscapes_stdc_seg50_dataset_params.yaml
@@ -24,24 +24,32 @@ train_dataset_params:
         crop_size: [ 1024, 512 ]
         mode: random
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 val_dataset_params:
   transforms:
     - SegRescale:
         scale_factor: 0.5
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 train_dataloader_params:
   batch_size: 16

diff --git a/src/super_gradients/recipes/dataset_params/cityscapes_stdc_seg75_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/cityscapes_stdc_seg75_dataset_params.yaml
@@ -24,23 +24,31 @@ train_dataset_params:
         crop_size: [ 1536, 768 ]
         mode: random
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 val_dataset_params:
   transforms:
     - SegRescale:
         scale_factor: 0.75
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
-        mean: [0.485, 0.456, 0.406]
-        std: [0.229, 0.224, 0.225]
+        mean: [ 0.485, 0.456, 0.406 ]
+        std: [ 0.229, 0.224, 0.225 ]
+
+    - SegConvertToTensor
+
 
 
 train_dataloader_params:

diff --git a/src/super_gradients/recipes/dataset_params/coco_segmentation_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/coco_segmentation_dataset_params.yaml
@@ -25,12 +25,16 @@ train_dataset_params:
         crop_size: 512
         mode: random
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 val_dataset_params:
   root_dir: /data/coco/
@@ -49,12 +53,16 @@ val_dataset_params:
         crop_size: 512
         mode: center
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 train_dataloader_params:
   shuffle: True

diff --git a/src/super_gradients/recipes/dataset_params/mapillary_dataset_params.yaml b/src/super_gradients/recipes/dataset_params/mapillary_dataset_params.yaml
@@ -36,12 +36,16 @@ train_dataset_params:
         crop_size: 1024
         mode: random
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 val_dataset_params:
   root_dir: /data/mapillary-vistas-dataset_public_v2.0
@@ -58,12 +62,16 @@ val_dataset_params:
         divisible_value: 32
         fill_mask: ${dataset_params.mask_fill_value}
 
-    - SegToTensor
+    - SegStandardize:
+        max_value: 255
 
     - SegNormalize:
         mean: [0.485, 0.456, 0.406]
         std: [0.229, 0.224, 0.225]
 
+    - SegConvertToTensor
+
+
 
 train_dataloader_params:
   shuffle: True