tests/testthat/test-models-resnet.R

In mlverse/torchvision: Models, Datasets and Transformations for Images

test_that("resnet18", {

  model <- model_resnet18()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  model <- model_resnet18(pretrained = TRUE)
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("resnet34", {
  skip_on_os(c("windows", "mac"))

  model <- model_resnet34()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  model <- model_resnet34(pretrained = TRUE)
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("resnet50", {
  skip_on_os(c("windows", "mac"))

  model <- model_resnet50()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  model <- model_resnet50(pretrained = TRUE)
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("resnet101", {
  skip_on_os(c("windows", "mac"))

  model <- model_resnet101()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  withr::with_options(list(timeout = 360),
                      model <- model_resnet101(pretrained = TRUE))
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("resnet152", {
  skip_on_os(c("windows", "mac"))

  model <- model_resnet152()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  withr::with_options(list(timeout = 360),
                      model <- model_resnet152(pretrained = TRUE))
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("resnext50_32x4d", {
  skip_on_os(c("windows", "mac"))

  model <- model_resnext50_32x4d()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  withr::with_options(list(timeout = 360),
                      model <- model_resnext50_32x4d(pretrained = TRUE))
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("resnext50_32x4d", {
  skip_on_os(c("windows", "mac"))

  model <- model_resnext50_32x4d()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  withr::with_options(list(timeout = 360),
                      model <- model_resnext50_32x4d(pretrained = TRUE))
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("resnext101_32x8d", {
  skip_on_os(c("windows", "mac"))

  model <- model_resnext101_32x8d()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  withr::with_options(list(timeout = 360),
                      model <- model_resnext101_32x8d(pretrained = TRUE))
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("wide_resnet50_2", {
  skip_on_os(c("windows", "mac"))

  model <- model_wide_resnet50_2()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  withr::with_options(list(timeout = 360),
                      model <- model_wide_resnet50_2(pretrained = TRUE))
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("wide_resnet101_2", {
  skip_on_os(c("windows", "mac"))

  model <- model_wide_resnet101_2()
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

  withr::with_options(list(timeout = 360),
                      model <- model_wide_resnet101_2(pretrained = TRUE))
  input <- torch::torch_randn(1, 3, 256, 256)
  out <- model(input)

  expect_tensor_shape(out, c(1, 1000))

})

test_that("we can prune head of resnet34 moels", {
  resnet34 <- model_resnet34(pretrained=TRUE)

  expect_error(prune <- nn_prune_head(resnet34, 1), NA)
  # expect_true(inherits(prune, "nn_sequential"))
  expect_equal(length(prune), 9)
  expect_true(inherits(prune[[length(prune)]], "nn_adaptive_avg_pool2d"))

  input <- torch::torch_randn(1, 3, 256, 256)
  out <- prune(input)
  expect_tensor_shape(out, c(1, 512, 1, 1))

})

test_that("we can prune head of resnet50 moels", {
  resnet50 <- model_resnet50(pretrained=TRUE)

  expect_error(prune <- nn_prune_head(resnet50, 1), NA)
  expect_true(inherits(prune, "nn_sequential"))
  expect_equal(length(prune), 9)
  expect_true(inherits(prune[[length(prune)]], "nn_adaptive_avg_pool2d"))

  input <- torch::torch_randn(1, 3, 256, 256)
  out <- prune(input)
  expect_tensor_shape(out, c(1, 2048, 1, 1))


})

test_that("we can prune head of resnext101 moels", {
  resnext101 <- model_resnext101_32x8d(pretrained=TRUE)

  expect_error(prune <- torch:::nn_prune_head(resnext101, 1), NA)
  expect_true(inherits(prune, "nn_sequential"))
  expect_equal(length(prune), 9)
  expect_true(inherits(prune[[length(prune)]], "nn_adaptive_avg_pool2d"))

  input <- torch::torch_randn(1, 3, 256, 256)
  out <- prune(input)
  expect_tensor_shape(out, c(1, 2048, 1, 1))


})