Faster R-CNN model
Your next task is to build a Faster R-CNN model that can detect objects of different sizes in an image. For this task, you will be using a handy class MultiScaleRoIAlign()
from torchvision.ops
.
FasterRCNN
class has been imported from torchvision.models.detection
. Your anchor_generator
from the last exercise is available in your workspace and torch
, torch.nn
as nn
, and torchvision
have been imported.
This exercise is part of the course
Deep Learning for Images with PyTorch
Exercise instructions
- Import
MultiScaleRoIAlign
fromtorchvision.ops
. - Instantiate the RoI pooler using
MultiScaleRoIAlign
withfeatmap_names
set to["0"]
,output_size
to7
, andsampling_ratio
to2
. - Create the Faster R-CNN model passing it the
backbone
,num_class
for a binary classification,anchor_generator
, androi_pooler
.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
# Import MultiScaleRoIAlign
____
# Instantiate RoI pooler
roi_pooler = ____(
____,
____,
____,
)
mobilenet = torchvision.models.mobilenet_v2(weights="DEFAULT")
backbone = nn.Sequential(*list(mobilenet.features.children()))
backbone.out_channels = 1280
# Create Faster R-CNN model
model = ____(
backbone=____
num_classes=____,
anchor_generator=____,
box_roi_pool=____,
)