Merge branch 'Nota-NetsPresso:dev' into 376-yolofastest

Nota-NetsPresso · Jun 25, 2024 · f16c38a · f16c38a
2 parents 5869c1d + 425c91c
commit f16c38a
Show file tree

Hide file tree

Showing 45 changed files with 170 additions and 79 deletions.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -3,6 +3,7 @@
 ## New Features:
 
 - Add YOLOX-nano and YOLOX-tiny by `@hglee98` in [PR 467](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/467)
+- Separate postprocessor configuration hierarchy by `@hglee98` in [PR 470](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/470)
 
 ## Bug Fixes:
 
@@ -15,6 +16,7 @@ No changes to highlight.
 ## Other Changes:
 
 - Change attention bias interpolate method by `@illian01` in [PR 468](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/468)
+- Fix ViT token number in positional encoding for torch.fx compile step by `@illian01` in [PR 475](https://github.com/Nota-NetsPresso/netspresso-trainer/pull/475)
 
 # v0.2.2
 

diff --git a/config/benchmark_examples/classification-imagenet1k-resnet18/model.yaml b/config/benchmark_examples/classification-imagenet1k-resnet18/model.yaml
@@ -38,6 +38,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20 
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.

diff --git a/config/benchmark_examples/classification-imagenet1k-resnet34/model.yaml b/config/benchmark_examples/classification-imagenet1k-resnet34/model.yaml
@@ -38,6 +38,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20 
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.

diff --git a/config/benchmark_examples/classification-imagenet1k-resnet50/model.yaml b/config/benchmark_examples/classification-imagenet1k-resnet50/model.yaml
@@ -38,6 +38,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20 
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/benchmark_examples/detection-coco2017-yolox_s/model.yaml b/config/benchmark_examples/detection-coco2017-yolox_s/model.yaml
@@ -26,11 +26,13 @@ model:
       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
+  postprocessor: 
+    params: 
+      # postprocessor - decode
+      score_thresh: 0.01
+      # postprocessor - nms
+      nms_thresh: 0.65
+      class_agnostic: False
   losses:
     - criterion: yolox_loss
       weight: ~

diff --git a/config/model/efficientformer/efficientformer-l1-classification.yaml b/config/model/efficientformer/efficientformer-l1-classification.yaml
@@ -41,6 +41,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/efficientformer/efficientformer-l1-detection.yaml b/config/model/efficientformer/efficientformer-l1-detection.yaml
@@ -48,13 +48,15 @@ model:
         anchor_sizes: [[32,], [64,], [128,], [256,]]
         aspect_ratios: [0.5, 1.0, 2.0]
         num_layers: 1
-        norm_type: batch_norm
-        # postprocessor - decode
-        topk_candidates: 1000
-        score_thresh: 0.05
-        # postprocessor - nms
-        nms_thresh: 0.45
-        class_agnostic: False
+        norm_type: batch_norm   
+  postprocessor: 
+    params:
+      # postprocessor - decode
+      topk_candidates: 1000
+      score_thresh: 0.05
+      # postprocessor - nms
+      nms_thresh: 0.45
+      class_agnostic: False
   losses:
     - criterion: retinanet_loss
       weight: ~
diff --git a/config/model/efficientformer/efficientformer-l1-segmentation.yaml b/config/model/efficientformer/efficientformer-l1-segmentation.yaml
@@ -39,6 +39,7 @@ model:
       params:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
+  postprocessor: ~
   losses:
     - criterion: seg_cross_entropy
       weight: ~

diff --git a/config/model/mixnet/mixnet-l-classification.yaml b/config/model/mixnet/mixnet-l-classification.yaml
@@ -65,6 +65,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/mixnet/mixnet-l-detection.yaml b/config/model/mixnet/mixnet-l-detection.yaml
@@ -73,12 +73,14 @@ model:
         aspect_ratios: [0.5, 1.0, 2.0]
         num_layers: 1
         norm_type: batch_norm
-        # postprocessor - decode
-        topk_candidates: 1000
-        score_thresh: 0.05
-        # postprocessor - nms
-        nms_thresh: 0.45
-        class_agnostic: False
+  postprocessor: 
+    params:
+      # postprocessor - decode
+      topk_candidates: 1000
+      score_thresh: 0.05
+      # postprocessor - nms
+      nms_thresh: 0.45
+      class_agnostic: False
   losses:
     - criterion: retinanet_loss
       weight: ~
diff --git a/config/model/mixnet/mixnet-l-segmentation.yaml b/config/model/mixnet/mixnet-l-segmentation.yaml
@@ -63,6 +63,7 @@ model:
       params:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
+  postprocessor: ~
   losses:
     - criterion: seg_cross_entropy
       weight: ~

diff --git a/config/model/mixnet/mixnet-m-classification.yaml b/config/model/mixnet/mixnet-m-classification.yaml
@@ -65,6 +65,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/mixnet/mixnet-m-detection.yaml b/config/model/mixnet/mixnet-m-detection.yaml
@@ -73,12 +73,14 @@ model:
         aspect_ratios: [0.5, 1.0, 2.0]
         num_layers: 1
         norm_type: batch_norm
-        # postprocessor - decode
-        topk_candidates: 1000
-        score_thresh: 0.05
-        # postprocessor - nms
-        nms_thresh: 0.45
-        class_agnostic: False
+  postprocessor: 
+    params:
+      # postprocessor - decode
+      topk_candidates: 1000
+      score_thresh: 0.05
+      # postprocessor - nms
+      nms_thresh: 0.45
+      class_agnostic: False
   losses:
     - criterion: retinanet_loss
       weight: ~
diff --git a/config/model/mixnet/mixnet-m-segmentation.yaml b/config/model/mixnet/mixnet-m-segmentation.yaml
@@ -63,6 +63,7 @@ model:
       params:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
+  postprocessor: ~
   losses:
     - criterion: seg_cross_entropy
       weight: ~

diff --git a/config/model/mixnet/mixnet-s-classification.yaml b/config/model/mixnet/mixnet-s-classification.yaml
@@ -65,6 +65,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/mixnet/mixnet-s-detection.yaml b/config/model/mixnet/mixnet-s-detection.yaml
@@ -73,12 +73,14 @@ model:
         aspect_ratios: [0.5, 1.0, 2.0]
         num_layers: 1
         norm_type: batch_norm
-        # postprocessor - decode
-        topk_candidates: 1000
-        score_thresh: 0.05
-        # postprocessor - nms
-        nms_thresh: 0.45
-        class_agnostic: False
+  postprocessor: 
+    params:
+      # postprocessor - decode
+      topk_candidates: 1000
+      score_thresh: 0.05
+      # postprocessor - nms
+      nms_thresh: 0.45
+      class_agnostic: False
   losses:
     - criterion: retinanet_loss
       weight: ~
diff --git a/config/model/mixnet/mixnet-s-segmentation.yaml b/config/model/mixnet/mixnet-s-segmentation.yaml
@@ -63,6 +63,7 @@ model:
       params:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
+  postprocessor: ~
   losses:
     - criterion: seg_cross_entropy
       weight: ~

diff --git a/config/model/mobilenetv3/mobilenetv3-large-classification.yaml b/config/model/mobilenetv3/mobilenetv3-large-classification.yaml
@@ -53,6 +53,9 @@ model:
         intermediate_channels: 1280
         act_type: hard_swish
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/mobilenetv3/mobilenetv3-small-classification.yaml b/config/model/mobilenetv3/mobilenetv3-small-classification.yaml
@@ -53,6 +53,9 @@ model:
         intermediate_channels: 1024
         act_type: hard_swish
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/mobilenetv3/mobilenetv3-small-detection.yaml b/config/model/mobilenetv3/mobilenetv3-small-detection.yaml
@@ -61,12 +61,14 @@ model:
         aspect_ratios: [0.5, 1.0, 2.0]
         num_layers: 1
         norm_type: batch_norm
-        # postprocessor - decode
-        topk_candidates: 1000
-        score_thresh: 0.05
-        # postprocessor - nms
-        nms_thresh: 0.45
-        class_agnostic: False
+  postprocessor: 
+    params:
+      # postprocessor - decode
+      topk_candidates: 1000
+      score_thresh: 0.05
+      # postprocessor - nms
+      nms_thresh: 0.45
+      class_agnostic: False
   losses:
     - criterion: retinanet_loss
       weight: ~
diff --git a/config/model/mobilenetv3/mobilenetv3-small-segmentation.yaml b/config/model/mobilenetv3/mobilenetv3-small-segmentation.yaml
@@ -51,6 +51,7 @@ model:
       params:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
+  postprocessor: ~
   losses:
     - criterion: seg_cross_entropy
       weight: ~

diff --git a/config/model/mobilevit/mobilevit-s-classification.yaml b/config/model/mobilevit/mobilevit-s-classification.yaml
@@ -66,6 +66,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/pidnet/pidnet-s-segmentation.yaml b/config/model/pidnet/pidnet-s-segmentation.yaml
@@ -18,6 +18,7 @@ model:
       head_channels: 128
     backbone: ~
     head: ~
+  postprocessor: ~
   losses:
     - criterion: pidnet_loss
       weight: ~

diff --git a/config/model/resnet/resnet18-classification.yaml b/config/model/resnet/resnet18-classification.yaml
@@ -38,6 +38,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/resnet/resnet34-classification.yaml b/config/model/resnet/resnet34-classification.yaml
@@ -38,6 +38,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/resnet/resnet50-classification.yaml b/config/model/resnet/resnet50-classification.yaml
@@ -38,6 +38,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1

diff --git a/config/model/resnet/resnet50-detection.yaml b/config/model/resnet/resnet50-detection.yaml
@@ -46,12 +46,14 @@ model:
         aspect_ratios: [0.5, 1.0, 2.0]
         num_layers: 1
         norm_type: batch_norm
-        # postprocessor - decode
-        topk_candidates: 1000
-        score_thresh: 0.05
-        # postprocessor - nms
-        nms_thresh: 0.45
-        class_agnostic: False
+  postprocessor: 
+    params:
+      # postprocessor - decode
+      topk_candidates: 1000
+      score_thresh: 0.05
+      # postprocessor - nms
+      nms_thresh: 0.45
+      class_agnostic: False
   losses:
     - criterion: retinanet_loss
       weight: ~
diff --git a/config/model/resnet/resnet50-segmentation.yaml b/config/model/resnet/resnet50-segmentation.yaml
@@ -37,6 +37,7 @@ model:
       params:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
+  postprocessor: ~
   losses:
     - criterion: seg_cross_entropy
       weight: ~

diff --git a/config/model/rtmpose/rtmpose-pose_estimation.yaml b/config/model/rtmpose/rtmpose-pose_estimation.yaml
@@ -62,6 +62,9 @@ model:
         simcc_split_ratio: 2.
         target_size: [256, 256]
         backbone_stride: 32
+  postprocessor: 
+    params: 
+      simcc_split_ratio: 2.
   losses:
     - criterion: rtmcc_loss
       weight: ~
diff --git a/config/model/segformer/segformer-b0-segmentation.yaml b/config/model/segformer/segformer-b0-segmentation.yaml
@@ -52,6 +52,7 @@ model:
         intermediate_channels: 256
         classifier_dropout_prob: 0.
         resize_output: [512, 512]
+  postprocessor: ~
   losses:
     - criterion: seg_cross_entropy
       weight: ~

diff --git a/config/model/vit/vit-tiny-classification.yaml b/config/model/vit/vit-tiny-classification.yaml
@@ -30,6 +30,9 @@ model:
         intermediate_channels: ~
         act_type: ~
         dropout_prob: 0.
+  postprocessor: 
+    params: 
+      topk_max: 20
   losses:
     - criterion: cross_entropy
       label_smoothing: 0.1