Merge pull request #462 from Nota-NetsPresso/v0.2.2-release

Micro update: v0.2.2 release
Nota-NetsPresso · Jun 11, 2024 · 6c4741e · 6c4741e
2 parents 3751061 + 996cf5d
commit 6c4741e
Show file tree

Hide file tree

Showing 206 changed files with 5,071 additions and 410 deletions.
diff --git a/ACKNOWLEDGEMENTS b/ACKNOWLEDGEMENTS
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -16,6 +16,28 @@ No changes to highlight.
 
 No changes to highlight.
 
+# v0.2.2
+
+## New Features:
+
+- Update Benchmarks & Checkpoints (docs) and weights files to fully usable by `@illian01` in [PR 446](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/446), [PR 447](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/447), [PR 456](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/456), [PR 461](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/461)
+- Add TFLite runtime code example by `@illian01` in [PR 449](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/449)
+
+## Bug Fixes:
+
+- Fix best_epoch init error in TrainingSummary in case of training resume by `@illian01` in [PR 448](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/448)
+- Fix segmentation metric logic bug by `@illian01` in [PR 455](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/455), [PR 460](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/460)
+
+## Breaking Changes:
+
+No changes to highlight.
+
+## Other Changes:
+
+- Refactoring: remove thop, replace MACs with FLOPs by `@illian01` in [PR 444](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/444)
+- Add copyright for entire project by `@illian01` in [PR 451](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/451)
+- Modify ImageSaver to receive various resolution at once `@illian01` in [PR 458](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/458)
+
 # v0.2.1
 
 ## New Features:

diff --git a/LICENSE b/LICENSE
@@ -199,3 +199,11 @@
    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    See the License for the specific language governing permissions and
    limitations under the License.
+
+
+------------------------------------------------------------------------------
+SOFTWARE LICENSE ACKNOWLEDGEMETS:
+
+This software utilizes several external components with separate copyright and license terms.
+Please see the ACKNOWLEDGEMENTS file.
+------------------------------------------------------------------------------
diff --git a/README.md b/README.md
@@ -113,6 +113,8 @@ If you are interested in utilizing open datasets, you can use them by following
 
 ### Semantic segmentation
 
+- [ADE20K](https://github.com/Nota-NetsPresso/netspresso-trainer/blob/dev/tools/open_dataset_tool/ade20k.py)
+- [Cityscapes](https://github.com/Nota-NetsPresso/netspresso-trainer/blob/dev/tools/open_dataset_tool/cityscapes.py)
 - [PascalVOC 2012](https://github.com/Nota-NetsPresso/netspresso-trainer/blob/dev/tools/open_dataset_tool/voc2012_seg.py)
 
 ### Object detection

diff --git a/config/benchmark_examples/detection-coco2017-yolox_s/augmentation.yaml b/config/benchmark_examples/detection-coco2017-yolox_s/augmentation.yaml
@@ -40,7 +40,7 @@ augmentation:
   inference:
     - 
       name: resize
-      size: 64
+      size: 640
       interpolation: bilinear
       max_size: ~
       resize_criteria: long

diff --git a/...data/local/voc12_grayscale_converted.yaml → config/data/local/ade20k.yaml b/...data/local/voc12_grayscale_converted.yaml → config/data/local/ade20k.yaml
@@ -1,21 +1,20 @@
 data:
-  name: voc2012
+  name: ade20k
   task: segmentation
   format: local
   path:
-    root: ./data/voc2012_seg
+    root: ./data/ade20k
     train:
       image: images/train # directory for training images
       label: labels/train # directory for training labels
     valid:
       image: images/valid # directory for valid images
       label: labels/valid # directory for valid labels
     test:
-      image: ~
-      label: ~
+      image: images/valid # directory for test images
+      label: labels/valid # directory for test images
     pattern:
       image: ~
       label: ~
   label_image_mode: L
   id_mapping: id_mapping.json
-  #id_mapping: ['background', 'aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus', 'car', 'cat', 'chair', 'cow', 'diningtable', 'dog', 'horse', 'motorbike', 'person', 'pottedplant', 'sheep', 'sofa', 'train', 'tvmonitor']
diff --git a/config/data/local/cityscapes.yaml b/config/data/local/cityscapes.yaml
@@ -0,0 +1,20 @@
+data:
+  name: cityscapes
+  task: segmentation
+  format: local
+  path:
+    root: ./data/cityscapes
+    train:
+      image: images/train # directory for training images
+      label: labels/train # directory for training labels
+    valid:
+      image: images/valid # directory for valid images
+      label: labels/valid # directory for valid labels
+    test:
+      image: images/valid # directory for test images
+      label: labels/valid # directory for test images
+    pattern:
+      image: ~
+      label: ~
+  label_image_mode: L
+  id_mapping: id_mapping.json
diff --git a/config/data/local/voc12.yaml b/config/data/local/voc12.yaml
@@ -3,7 +3,7 @@ data:
   task: segmentation
   format: local
   path:
-    root: ./data/voc2012_seg # dataset root
+    root: ./data/voc2012_seg
     train:
       image: images/train # directory for training images
       label: labels/train # directory for training labels
@@ -16,28 +16,6 @@ data:
     pattern:
       image: ~
       label: ~
-  label_image_mode: RGB
+  label_image_mode: L
   id_mapping: id_mapping.json
-  #   (0, 0, 0): background
-  #   (128, 0, 0): aeroplane
-  #   (0, 128, 0): bicycle
-  #   (128, 128, 0): bird
-  #   (0, 0, 128): boat
-  #   (128, 0, 128): bottle
-  #   (0, 128, 128): bus
-  #   (128, 128, 128): car
-  #   (64, 0, 0): cat
-  #   (192, 0, 0): chair
-  #   (64, 128, 0): cow
-  #   (192, 128, 0): diningtable
-  #   (64, 0, 128): dog
-  #   (192, 0, 128): horse
-  #   (64, 128, 128): motorbike
-  #   (192, 128, 128): person
-  #   (0, 64, 0): pottedplant
-  #   (128, 64, 0): sheep
-  #   (0, 192, 0): sofa
-  #   (128, 192, 0): train
-  #   (0, 64, 128): tvmonitor
-  #   (128, 64, 128): void
-  pallete: ~
+  #id_mapping: ['background', 'aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus', 'car', 'cat', 'chair', 'cow', 'diningtable', 'dog', 'horse', 'motorbike', 'person', 'pottedplant', 'sheep', 'sofa', 'train', 'tvmonitor']
diff --git a/config/data/local/voc12_rgb_mask.yaml b/config/data/local/voc12_rgb_mask.yaml
@@ -0,0 +1,42 @@
+data:
+  name: voc2012
+  task: segmentation
+  format: local
+  path:
+    root: ./data/voc2012_seg # dataset root
+    train:
+      image: images/train # directory for training images
+      label: labels/train # directory for training labels
+    valid:
+      image: images/valid # directory for valid images
+      label: labels/valid # directory for valid labels
+    test:
+      image: ~
+      label: ~
+    pattern:
+      image: ~
+      label: ~
+  label_image_mode: RGB
+  id_mapping: id_mapping.json
+  #   (0, 0, 0): background
+  #   (128, 0, 0): aeroplane
+  #   (0, 128, 0): bicycle
+  #   (128, 128, 0): bird
+  #   (0, 0, 128): boat
+  #   (128, 0, 128): bottle
+  #   (0, 128, 128): bus
+  #   (128, 128, 128): car
+  #   (64, 0, 0): cat
+  #   (192, 0, 0): chair
+  #   (64, 128, 0): cow
+  #   (192, 128, 0): diningtable
+  #   (64, 0, 128): dog
+  #   (192, 0, 128): horse
+  #   (64, 128, 128): motorbike
+  #   (192, 128, 128): person
+  #   (0, 64, 0): pottedplant
+  #   (128, 64, 0): sheep
+  #   (0, 192, 0): sofa
+  #   (128, 192, 0): train
+  #   (0, 64, 128): tvmonitor
+  pallete: ~
diff --git a/config/model/efficientformer/efficientformer-l1-segmentation.yaml b/config/model/efficientformer/efficientformer-l1-segmentation.yaml
@@ -40,6 +40,6 @@ model:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
   losses:
-    - criterion: cross_entropy
+    - criterion: seg_cross_entropy
       weight: ~
-      ignore_index: 255
+      ignore_index: 255 # Default undefined class is 255
diff --git a/config/model/mixnet/mixnet-l-segmentation.yaml b/config/model/mixnet/mixnet-l-segmentation.yaml
@@ -64,6 +64,6 @@ model:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
   losses:
-    - criterion: cross_entropy
+    - criterion: seg_cross_entropy
       weight: ~
-      ignore_index: 255
+      ignore_index: 255 # Default undefined class is 255
diff --git a/config/model/mixnet/mixnet-m-segmentation.yaml b/config/model/mixnet/mixnet-m-segmentation.yaml
@@ -64,6 +64,6 @@ model:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
   losses:
-    - criterion: cross_entropy
+    - criterion: seg_cross_entropy
       weight: ~
-      ignore_index: 255
+      ignore_index: 255 # Default undefined class is 255
diff --git a/config/model/mixnet/mixnet-s-segmentation.yaml b/config/model/mixnet/mixnet-s-segmentation.yaml
@@ -64,6 +64,6 @@ model:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
   losses:
-    - criterion: cross_entropy
+    - criterion: seg_cross_entropy
       weight: ~
-      ignore_index: 255
+      ignore_index: 255 # Default undefined class is 255
diff --git a/config/model/mobilenetv3/mobilenetv3-large-classification.yaml b/config/model/mobilenetv3/mobilenetv3-large-classification.yaml
@@ -0,0 +1,59 @@
+model:
+  task: classification
+  name: mobilenet_v3_large
+  checkpoint:
+    use_pretrained: True
+    load_head: False
+    path: ~
+    fx_model_path: ~
+    optimizer_path: ~
+  freeze_backbone: False
+  architecture:
+    full: ~ # auto
+    backbone:
+      name: mobilenetv3
+      params: ~
+      stage_params:
+        -
+          in_channels: [16, 16, 24]
+          kernel_sizes: [3, 3, 3]
+          expanded_channels: [16, 64, 72]
+          out_channels: [16, 24, 24]
+          use_se: [False, False, False]
+          act_type: ["relu", "relu", "relu"]
+          stride: [1, 2, 1]
+        -
+          in_channels: [24, 40, 40]
+          kernel_sizes: [5, 5, 5]
+          expanded_channels: [72, 120, 120]
+          out_channels: [40, 40, 40]
+          use_se: [True, True, True]
+          act_type: ["relu", "relu", "relu"]
+          stride: [2, 1, 1]
+        -
+          in_channels: [40, 80, 80, 80, 80, 112]
+          kernel_sizes: [3, 3, 3, 3, 3, 3]
+          expanded_channels: [240, 200, 184, 184, 480, 672]
+          out_channels: [80, 80, 80, 80, 112, 112]
+          use_se: [False, False, False, False, True, True]
+          act_type: ["hard_swish", "hard_swish", "hard_swish", "hard_swish", "hard_swish", "hard_swish"]
+          stride: [2, 1, 1, 1, 1, 1]
+        -
+          in_channels: [112, 160, 160]
+          kernel_sizes: [5, 5, 5]
+          expanded_channels: [672, 960, 960]
+          out_channels: [160, 160, 160]
+          use_se: [True, True, True]
+          act_type: ["hard_swish", "hard_swish", "hard_swish"]
+          stride: [2, 1, 1]
+    head:
+      name: fc
+      params:
+        num_layers: 2
+        intermediate_channels: 1280
+        act_type: hard_swish
+        dropout_prob: 0.
+  losses:
+    - criterion: cross_entropy
+      label_smoothing: 0.1
+      weight: ~
diff --git a/config/model/mobilenetv3/mobilenetv3-small-classification.yaml b/config/model/mobilenetv3/mobilenetv3-small-classification.yaml
@@ -49,9 +49,9 @@ model:
     head:
       name: fc
       params:
-        num_layers: 1
-        intermediate_channels: ~
-        act_type: ~
+        num_layers: 2
+        intermediate_channels: 1024
+        act_type: hard_swish
         dropout_prob: 0.
   losses:
     - criterion: cross_entropy

diff --git a/config/model/mobilenetv3/mobilenetv3-small-segmentation.yaml b/config/model/mobilenetv3/mobilenetv3-small-segmentation.yaml
@@ -52,6 +52,6 @@ model:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
   losses:
-    - criterion: cross_entropy
+    - criterion: seg_cross_entropy
       weight: ~
-      ignore_index: 255
+      ignore_index: 255 # Default undefined class is 255
diff --git a/config/model/resnet/resnet50-segmentation.yaml b/config/model/resnet/resnet50-segmentation.yaml
@@ -38,6 +38,6 @@ model:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
   losses:
-    - criterion: cross_entropy
+    - criterion: seg_cross_entropy
       weight: ~
-      ignore_index: 255
+      ignore_index: 255 # Default undefined class is 255
diff --git a/config/model/segformer/segformer-b0-segmentation.yaml b/config/model/segformer/segformer-b0-segmentation.yaml
@@ -53,6 +53,6 @@ model:
         classifier_dropout_prob: 0.
         resize_output: [512, 512]
   losses:
-    - criterion: cross_entropy
+    - criterion: seg_cross_entropy
       weight: ~
-      ignore_index: 255
+      ignore_index: 255 # Default undefined class is 255
diff --git a/config/model/yolox/yolox-l-detection.yaml b/config/model/yolox/yolox-l-detection.yaml
@@ -0,0 +1,37 @@
+model:
+  task: detection
+  name: yolox_l
+  checkpoint:
+    use_pretrained: True
+    load_head: False
+    path: ~
+    fx_model_path: ~
+    optimizer_path: ~
+  freeze_backbone: False
+  architecture:
+    full: ~ # auto
+    backbone:
+      name: cspdarknet
+      params:
+        dep_mul: &dep_mul 1.0
+        wid_mul: 1.0
+        act_type: &act_type "silu"
+      stage_params: ~
+    neck:
+      name: yolopafpn
+      params:
+        dep_mul: *dep_mul
+        act_type: *act_type
+    head:
+      name: anchor_free_decoupled_head
+      params:
+        act_type: *act_type
+        # postprocessor - decode
+        score_thresh: 0.01
+        # postprocessor - nms
+        nms_thresh: 0.65
+        class_agnostic: False
+  losses:
+    - criterion: yolox_loss
+      weight: ~
+      l1_activate_epoch: 1