Compile PyTorch Object Detection Models

This article is an introductory tutorial to deploy PyTorch object detection models with Relay VM.

For us to begin with, PyTorch should be installed. TorchVision is also required since we will be using it as our model zoo.

A quick solution is to install via pip

pip install torch==1.7.0
pip install torchvision==0.8.1

or please refer to official site https://pytorch.org/get-started/locally/

PyTorch versions should be backwards compatible but should be used with the proper TorchVision version.

Currently, TVM supports PyTorch 1.7 and 1.4. Other versions may be unstable.

import tvm
from tvm import relay
from tvm import relay
from tvm.runtime.vm import VirtualMachine
from tvm.contrib.download import download_testdata

import numpy as np
import cv2

# PyTorch imports
import torch
import torchvision

Load pre-trained maskrcnn from torchvision and do tracing

in_size = 300

input_shape = (1, 3, in_size, in_size)


def do_trace(model, inp):
    model_trace = torch.jit.trace(model, inp)
    model_trace.eval()
    return model_trace


def dict_to_tuple(out_dict):
    if "masks" in out_dict.keys():
        return out_dict["boxes"], out_dict["scores"], out_dict["labels"], out_dict["masks"]
    return out_dict["boxes"], out_dict["scores"], out_dict["labels"]


class TraceWrapper(torch.nn.Module):
    def __init__(self, model):
        super().__init__()
        self.model = model

    def forward(self, inp):
        out = self.model(inp)
        return dict_to_tuple(out[0])


model_func = torchvision.models.detection.maskrcnn_resnet50_fpn
model = TraceWrapper(model_func(pretrained=True))

model.eval()
inp = torch.Tensor(np.random.uniform(0.0, 250.0, size=(1, 3, in_size, in_size)))

with torch.no_grad():
    out = model(inp)
    script_module = do_trace(model, inp)

Downloading: "https://download.pytorch.org/models/maskrcnn_resnet50_fpn_coco-bf2d0c1e.pth" to /workspace/.cache/torch/hub/checkpoints/maskrcnn_resnet50_fpn_coco-bf2d0c1e.pth

  0%|          | 0.00/170M [00:00<?, ?B/s]
  2%|1         | 2.94M/170M [00:00<00:05, 30.4MB/s]
  4%|4         | 7.18M/170M [00:00<00:04, 38.6MB/s]
  7%|7         | 12.3M/170M [00:00<00:03, 45.7MB/s]
 10%|#         | 17.0M/170M [00:00<00:03, 46.8MB/s]
 13%|#2        | 21.5M/170M [00:00<00:03, 46.5MB/s]
 15%|#5        | 25.9M/170M [00:00<00:03, 43.3MB/s]
 18%|#7        | 30.1M/170M [00:00<00:03, 40.9MB/s]
 20%|##        | 34.0M/170M [00:00<00:03, 37.9MB/s]
 22%|##2       | 37.7M/170M [00:01<00:03, 34.7MB/s]
 25%|##5       | 43.2M/170M [00:01<00:03, 40.9MB/s]
 28%|##7       | 47.5M/170M [00:01<00:03, 41.8MB/s]
 30%|###       | 51.5M/170M [00:01<00:03, 36.5MB/s]
 32%|###2      | 55.2M/170M [00:01<00:03, 35.4MB/s]
 35%|###4      | 58.7M/170M [00:01<00:03, 30.2MB/s]
 36%|###6      | 61.9M/170M [00:01<00:03, 31.0MB/s]
 39%|###9      | 66.7M/170M [00:01<00:03, 35.8MB/s]
 42%|####1     | 71.2M/170M [00:01<00:02, 38.9MB/s]
 44%|####4     | 75.1M/170M [00:02<00:02, 39.0MB/s]
 46%|####6     | 78.9M/170M [00:02<00:02, 36.0MB/s]
 49%|####8     | 83.1M/170M [00:02<00:02, 37.9MB/s]
 51%|#####1    | 87.2M/170M [00:02<00:02, 39.0MB/s]
 54%|#####3    | 91.5M/170M [00:02<00:02, 41.0MB/s]
 56%|#####6    | 95.7M/170M [00:02<00:01, 41.4MB/s]
 59%|#####8    | 100M/170M [00:02<00:01, 42.9MB/s]
 61%|######1   | 104M/170M [00:02<00:01, 43.1MB/s]
 64%|######4   | 109M/170M [00:02<00:01, 44.8MB/s]
 67%|######6   | 113M/170M [00:03<00:01, 43.5MB/s]
 70%|######9   | 119M/170M [00:03<00:01, 47.8MB/s]
 73%|#######2  | 123M/170M [00:03<00:01, 47.8MB/s]
 75%|#######5  | 128M/170M [00:03<00:01, 43.5MB/s]
 78%|#######7  | 132M/170M [00:03<00:01, 33.6MB/s]
 80%|########  | 137M/170M [00:03<00:00, 36.2MB/s]
 83%|########3 | 141M/170M [00:03<00:00, 39.2MB/s]
 86%|########5 | 145M/170M [00:03<00:00, 39.5MB/s]
 88%|########7 | 149M/170M [00:03<00:00, 37.0MB/s]
 90%|######### | 154M/170M [00:04<00:00, 38.9MB/s]
 93%|#########2| 157M/170M [00:04<00:00, 36.0MB/s]
 95%|#########5| 161M/170M [00:04<00:00, 37.3MB/s]
 98%|#########7| 166M/170M [00:04<00:00, 39.4MB/s]
100%|#########9| 170M/170M [00:04<00:00, 38.5MB/s]
100%|##########| 170M/170M [00:04<00:00, 39.2MB/s]
/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py:3878: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  for i in range(dim)
/usr/local/lib/python3.7/dist-packages/torchvision/models/detection/anchor_utils.py:127: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
  for g in grid_sizes
/usr/local/lib/python3.7/dist-packages/torchvision/models/detection/anchor_utils.py:127: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  for g in grid_sizes
/usr/local/lib/python3.7/dist-packages/torchvision/models/detection/rpn.py:73: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
  A = Ax4 // 4
/usr/local/lib/python3.7/dist-packages/torchvision/models/detection/rpn.py:74: UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). This results in incorrect rounding for negative values. To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), or for actual floor division, use torch.div(a, b, rounding_mode='floor').
  C = AxC // A
/usr/local/lib/python3.7/dist-packages/torchvision/ops/boxes.py:156: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  boxes_x = torch.min(boxes_x, torch.tensor(width, dtype=boxes.dtype, device=boxes.device))
/usr/local/lib/python3.7/dist-packages/torchvision/ops/boxes.py:158: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  boxes_y = torch.min(boxes_y, torch.tensor(height, dtype=boxes.dtype, device=boxes.device))
/usr/local/lib/python3.7/dist-packages/torchvision/models/detection/transform.py:293: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  for s, s_orig in zip(new_size, original_size)
/usr/local/lib/python3.7/dist-packages/torchvision/models/detection/roi_heads.py:387: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  return torch.tensor(M + 2 * padding).to(torch.float32) / torch.tensor(M).to(torch.float32)

Download a test image and pre-process

img_url = (
    "https://raw.githubusercontent.com/dmlc/web-data/master/gluoncv/detection/street_small.jpg"
)
img_path = download_testdata(img_url, "test_street_small.jpg", module="data")

img = cv2.imread(img_path).astype("float32")
img = cv2.resize(img, (in_size, in_size))
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
img = np.transpose(img / 255.0, [2, 0, 1])
img = np.expand_dims(img, axis=0)

Import the graph to Relay

input_name = "input0"
shape_list = [(input_name, input_shape)]
mod, params = relay.frontend.from_pytorch(script_module, shape_list)

/workspace/python/tvm/relay/build_module.py:348: DeprecationWarning: Please use input parameter mod (tvm.IRModule) instead of deprecated parameter mod (tvm.relay.function.Function)
  DeprecationWarning,
/workspace/python/tvm/relay/frontend/pytorch.py:430: DeprecationWarning: `np.int` is a deprecated alias for the builtin `int`. To silence this warning, use `int` by itself. Doing this will not modify any behavior and is safe. When replacing `np.int`, you may wish to use e.g. `np.int64` or `np.int32` to specify the precision. If you wish to review your current use, check the release note link for additional information.
Deprecated in NumPy 1.20; for more details and guidance: https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations
  inputs[3], lambda ret: ret.astype(np.int).item(0)

Compile with Relay VM

Note: Currently only CPU target is supported. For x86 target, it is highly recommended to build TVM with Intel MKL and Intel OpenMP to get best performance, due to the existence of large dense operator in torchvision rcnn models.

# Add "-libs=mkl" to get best performance on x86 target.
# For x86 machine supports AVX512, the complete target is
# "llvm -mcpu=skylake-avx512 -libs=mkl"
target = "llvm"

with tvm.transform.PassContext(opt_level=3, disabled_pass=["FoldScaleAxis"]):
    vm_exec = relay.vm.compile(mod, target=target, params=params)

Inference with Relay VM

dev = tvm.cpu()
vm = VirtualMachine(vm_exec, dev)
vm.set_input("main", **{input_name: img})
tvm_res = vm.run()

Get boxes with score larger than 0.9

score_threshold = 0.9
boxes = tvm_res[0].numpy().tolist()
valid_boxes = []
for i, score in enumerate(tvm_res[1].numpy().tolist()):
    if score > score_threshold:
        valid_boxes.append(boxes[i])
    else:
        break

print("Get {} valid boxes".format(len(valid_boxes)))

Get 9 valid boxes