DETR Model Export

Summary:
Pull Request resolved: #169

Make d2go DETR exportable (torchscript compatible)
Move generating masks to preprocessing

Reviewed By: sstsai-adl

Differential Revision: D33798073

fbshipit-source-id: d629b0c9cbdb67060982be717c7138a0e7e9adbc

projects_oss/detr/detr/d2/detr.py

@@ -124,8 +124,10 @@ def forward(self, batched_inputs):
             dict[str: Tensor]:
                 mapping from a named loss to a tensor storing the loss. Used during training only.
         """
-        images = self.preprocess_image(batched_inputs)
-        output = self.detr(images)
+        images_lists = self.preprocess_image(batched_inputs)
+        # convert images_lists to Nested Tensor?
+        nested_images = self.imagelist_to_nestedtensor(images_lists)
+        output = self.detr(nested_images)
 
         if self.training:
             gt_instances = [x["instances"].to(self.device) for x in batched_inputs]
@@ -144,10 +146,12 @@ def forward(self, batched_inputs):
             box_cls = output["pred_logits"]
             box_pred = output["pred_boxes"]
             mask_pred = output["pred_masks"] if self.mask_on else None
-            results = self.inference(box_cls, box_pred, mask_pred, images.image_sizes)
+            results = self.inference(
+                box_cls, box_pred, mask_pred, images_lists.image_sizes
+            )
             processed_results = []
             for results_per_image, input_per_image, image_size in zip(
-                results, batched_inputs, images.image_sizes
+                results, batched_inputs, images_lists.image_sizes
             ):
                 height = input_per_image.get("height", image_size[0])
                 width = input_per_image.get("width", image_size[1])
@@ -239,3 +243,12 @@ def preprocess_image(self, batched_inputs):
         images = [self.normalizer(x["image"].to(self.device)) for x in batched_inputs]
         images = ImageList.from_tensors(images)
         return images
+
+    def imagelist_to_nestedtensor(self, images):
+        tensor = images.tensor
+        device = tensor.device
+        N, _, H, W = tensor.shape
+        masks = torch.ones((N, H, W), dtype=torch.bool, device=device)
+        for idx, (h, w) in enumerate(images.image_sizes):
+            masks[idx, :h, :w] = False
+        return NestedTensor(tensor, masks)

projects_oss/detr/detr/models/build.py

@@ -1,5 +1,8 @@
+from typing import Dict
+
 import numpy as np
 import torch
+import torch.nn.functional as F
 from detectron2.modeling import build_backbone
 from detectron2.utils.registry import Registry
 from detr.models.backbone import Joiner
@@ -59,35 +62,15 @@ def __init__(self, cfg):
         self.feature_strides = [backbone_shape[f].stride for f in backbone_shape.keys()]
         self.num_channels = [backbone_shape[k].channels for k in backbone_shape.keys()]
 
-    def forward(self, images):
-        features = self.backbone(images.tensor)
-        # one tensor per feature level. Each tensor has shape (B, maxH, maxW)
-        masks = self.mask_out_padding(
-            [features_per_level.shape for features_per_level in features.values()],
-            images.image_sizes,
-            images.tensor.device,
-        )
-        assert len(features) == len(masks)
-        for i, k in enumerate(features.keys()):
-            features[k] = NestedTensor(features[k], masks[i])
-        return features
-
-    def mask_out_padding(self, feature_shapes, image_sizes, device):
-        masks = []
-        assert len(feature_shapes) == len(self.feature_strides)
-        for idx, shape in enumerate(feature_shapes):
-            N, _, H, W = shape
-            masks_per_feature_level = torch.ones(
-                (N, H, W), dtype=torch.bool, device=device
-            )
-            for img_idx, (h, w) in enumerate(image_sizes):
-                masks_per_feature_level[
-                    img_idx,
-                    : int(np.ceil(float(h) / self.feature_strides[idx])),
-                    : int(np.ceil(float(w) / self.feature_strides[idx])),
-                ] = 0
-            masks.append(masks_per_feature_level)
-        return masks
+    def forward(self, tensor_list: NestedTensor):
+        xs = self.backbone(tensor_list.tensors)
+        out: Dict[str, NestedTensor] = {}
+        for name, x in xs.items():
+            m = tensor_list.mask
+            assert m is not None
+            mask = F.interpolate(m[None].float(), size=x.shape[-2:]).to(torch.bool)[0]
+            out[name] = NestedTensor(x, mask)
+        return out
 
 
 class FBNetMaskedBackbone(ResNetMaskedBackbone):
@@ -105,20 +88,6 @@ def __init__(self, cfg):
             self.backbone._out_feature_strides[k] for k in self.out_features
         ]
 
-    def forward(self, images):
-        features = self.backbone(images.tensor)
-        masks = self.mask_out_padding(
-            [features_per_level.shape for features_per_level in features.values()],
-            images.image_sizes,
-            images.tensor.device,
-        )
-        assert len(features) == len(masks)
-        ret_features = {}
-        for i, k in enumerate(features.keys()):
-            if k in self.out_features:
-                ret_features[k] = NestedTensor(features[k], masks[i])
-        return ret_features
-
 
 class SimpleSingleStageBackbone(ResNetMaskedBackbone):
     """This is a simple wrapper for single stage backbone,
@@ -135,15 +104,3 @@ def __init__(self, cfg):
         self.feature_strides = [cfg.MODEL.BACKBONE.STRIDE]
         self.num_channels = [cfg.MODEL.BACKBONE.CHANNEL]
         self.strides = [cfg.MODEL.BACKBONE.STRIDE]
-
-    def forward(self, images):
-        y = self.backbone(images.tensor)
-        masks = self.mask_out_padding(
-            [y.shape],
-            images.image_sizes,
-            images.tensor.device,
-        )
-        assert len(masks) == 1
-        ret_features = {}
-        ret_features[self.out_features[0]] = NestedTensor(y, masks[0])
-        return ret_features

projects_oss/detr/test_all.py

@@ -1,6 +1,5 @@
 #!/usr/bin/env python3
 # Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
-# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
 import io
 import unittest
 from typing import List

projects_oss/detr/test_detr_export.py

@@ -0,0 +1,67 @@
+#!/usr/bin/env python3
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+
+import unittest
+
+import torch
+from d2go.runner import create_runner
+from detr.util.misc import nested_tensor_from_tensor_list
+from fvcore.nn import flop_count_table, FlopCountAnalysis
+
+
+class Tester(unittest.TestCase):
+    @staticmethod
+    def _set_detr_cfg(cfg, enc_layers, dec_layers, num_queries, dim_feedforward):
+        cfg.MODEL.META_ARCHITECTURE = "Detr"
+        cfg.MODEL.DETR.NUM_OBJECT_QUERIES = num_queries
+        cfg.MODEL.DETR.ENC_LAYERS = enc_layers
+        cfg.MODEL.DETR.DEC_LAYERS = dec_layers
+        cfg.MODEL.DETR.DEEP_SUPERVISION = False
+        cfg.MODEL.DETR.DIM_FEEDFORWARD = dim_feedforward  # 2048
+
+    def _assert_model_output(self, model, scripted_model):
+        x = nested_tensor_from_tensor_list(
+            [torch.rand(3, 200, 200), torch.rand(3, 200, 250)]
+        )
+        out = model(x)
+        out_script = scripted_model(x)
+        self.assertTrue(out["pred_logits"].equal(out_script["pred_logits"]))
+        self.assertTrue(out["pred_boxes"].equal(out_script["pred_boxes"]))
+
+    def test_detr_res50_export(self):
+        runner = create_runner("d2go.projects.detr.runner.DETRRunner")
+        cfg = runner.get_default_cfg()
+        cfg.MODEL.DEVICE = "cpu"
+        # DETR
+        self._set_detr_cfg(cfg, 6, 6, 100, 2048)
+        # backbone
+        cfg.MODEL.BACKBONE.NAME = "build_resnet_backbone"
+        cfg.MODEL.RESNETS.DEPTH = 50
+        cfg.MODEL.RESNETS.STRIDE_IN_1X1 = False
+        cfg.MODEL.RESNETS.OUT_FEATURES = ["res2", "res3", "res4", "res5"]
+        # build model
+        model = runner.build_model(cfg).eval()
+        model = model.detr
+        scripted_model = torch.jit.script(model)
+        self._assert_model_output(model, scripted_model)
+
+    def test_detr_fbnet_export(self):
+        runner = create_runner("d2go.projects.detr.runner.DETRRunner")
+        cfg = runner.get_default_cfg()
+        cfg.MODEL.DEVICE = "cpu"
+        # DETR
+        self._set_detr_cfg(cfg, 3, 3, 50, 256)
+        # backbone
+        cfg.MODEL.BACKBONE.NAME = "FBNetV2C4Backbone"
+        cfg.MODEL.FBNET_V2.ARCH = "FBNetV3_A_dsmask_C5"
+        cfg.MODEL.FBNET_V2.WIDTH_DIVISOR = 8
+        cfg.MODEL.FBNET_V2.OUT_FEATURES = ["trunk4"]
+        # build model
+        model = runner.build_model(cfg).eval()
+        model = model.detr
+        print(model)
+        scripted_model = torch.jit.script(model)
+        self._assert_model_output(model, scripted_model)
+        # print flops
+        table = flop_count_table(FlopCountAnalysis(model, ([torch.rand(3, 224, 320)],)))
+        print(table)