tests/test_omorpher.py

"""
Tests for the OMorpher module.

Split into two groups:
  - Basic tests: verify shapes, value ranges, and API behaviour (no checkpoint needed)
  - Alignment tests: cross-validate against DeformDDPM / OM_reg_flexres (shared weights)

Run:
    python -m pytest tests/test_omorpher.py -v
    # or directly:
    python tests/test_omorpher.py
"""

import os
import sys
import math

import numpy as np
import torch
import torch.nn.functional as F

# Ensure project root is importable
ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
sys.path.insert(0, ROOT)

from OMorpher import OMorpher
from Diffusion.diffuser import DeformDDPM
from Diffusion.networks import get_net, STN

# ---------- helpers ----------

NDIMS = 3
IMG_SIZE = 32  # tiny for speed
TIMESTEPS = 10
NET_NAME = "recmutattnnet"
DEVICE = "cpu"

BASE_CONFIG = {
    "net_name": NET_NAME,
    "ndims": NDIMS,
    "img_size": IMG_SIZE,
    "timesteps": TIMESTEPS,
    "v_scale": 5e-5,
    "noise_scale": 0.1,
    "condition_type": "none",
    "num_input_chn": 1,
    "img_pad_mode": "zeros",
    "ddf_pad_mode": "border",
    "padding_mode": "border",
    "resample_mode": "bilinear",
    "batchsize": 1,
    "data_name": "test",
    "start_noise_step": 0,
}


def _make_omorpher(**overrides):
    cfg = {**BASE_CONFIG, **overrides}
    return OMorpher(config=cfg, checkpoint_path=None, device=DEVICE)


def _rand_vol(B=1, S=None):
    S = S or IMG_SIZE
    return torch.rand([B, 1] + [S] * NDIMS)


# ================================================================
# 1. Basic tests
# ================================================================

class TestInstantiation:
    """Test 1: OMorpher with config dict + no checkpoint."""

    def test_creates_network_and_stns(self):
        om = _make_omorpher()
        assert om.network is not None
        assert om.stn_full is not None
        assert om.stn_ctl is not None
        assert om.img_stn is not None
        assert om.msk_stn is not None

    def test_device(self):
        om = _make_omorpher()
        assert om.device == torch.device(DEVICE)

    def test_repr(self):
        om = _make_omorpher()
        r = repr(om)
        assert "OMorpher" in r
        assert NET_NAME in r


class TestStandardization:
    """Test 2: _standardize_img produces correct shape and range."""

    def test_numpy_input(self):
        om = _make_omorpher()
        vol = np.random.rand(40, 50, 60).astype(np.float32) * 1000.0
        tensor, fullres, orig_shape = om._standardize_img(vol, keep_raw=True)
        assert tensor.shape == (1, 1, IMG_SIZE, IMG_SIZE, IMG_SIZE)
        assert tensor.min() >= 0.0
        assert tensor.max() <= 1.0 + 1e-6
        assert fullres is not None
        assert isinstance(fullres, torch.Tensor)
        # fullres should be [1, 1, 60, 60, 60] (cube-padded from max dim)
        assert fullres.ndim == 5
        assert fullres.shape[2] == fullres.shape[3] == fullres.shape[4] == 60
        # orig_shape should be the cube-padded size
        assert orig_shape[0] == orig_shape[1] == orig_shape[2]

    def test_torch_passthrough(self):
        om = _make_omorpher()
        vol = torch.rand(1, 1, IMG_SIZE, IMG_SIZE, IMG_SIZE)
        tensor, raw, _ = om._standardize_img(vol)
        assert tensor.shape == vol.shape


class TestLabelStandardization:
    """Test: _standardize_label produces correct shapes and handles None."""

    def test_3d_label(self):
        om = _make_omorpher()
        om.set_init_img(_rand_vol().numpy()[0, 0])
        label = (np.random.rand(40, 50, 60) > 0.5).astype(np.float32)
        model_t, fullres_t = om._standardize_label(label)
        assert model_t.shape == (1, 1, IMG_SIZE, IMG_SIZE, IMG_SIZE)
        assert fullres_t.ndim == 5
        # cube-padded: max(40,50,60)=60
        assert fullres_t.shape[2] == fullres_t.shape[3] == fullres_t.shape[4] == 60
        assert isinstance(model_t, torch.Tensor)
        assert isinstance(fullres_t, torch.Tensor)

    def test_none_placeholder(self):
        om = _make_omorpher()
        om.set_init_img(_rand_vol().numpy()[0, 0])
        model_t, fullres_t = om._standardize_label(None)
        assert model_t.shape == (1, 1, IMG_SIZE, IMG_SIZE, IMG_SIZE)
        assert torch.all(model_t == -1)
        assert torch.all(fullres_t == -1)

    def test_4d_label(self):
        om = _make_omorpher()
        om.set_init_img(_rand_vol().numpy()[0, 0])
        label = (np.random.rand(30, 30, 30, 2) > 0.5).astype(np.float32)
        model_t, fullres_t = om._standardize_label(label)
        # 4D → channels-first: 2 channels
        assert model_t.shape == (1, 2, IMG_SIZE, IMG_SIZE, IMG_SIZE)
        assert fullres_t.shape[1] == 2


class TestZeroDDFRoundtrip:
    """Test 3: apply_def with zero DDF returns approx original."""

    def test_identity(self):
        om = _make_omorpher()
        vol = _rand_vol()
        zero_ddf = torch.zeros(1, NDIMS, IMG_SIZE, IMG_SIZE, IMG_SIZE)
        warped = om._apply_ddf(vol, zero_ddf, padding_mode="border")
        assert torch.allclose(vol, warped, atol=1e-5)


class TestSizeMismatch:
    """Test 4: apply_def auto-upscales DDF."""

    def test_upscale(self):
        om = _make_omorpher()
        big_vol = _rand_vol(S=64)
        small_ddf = torch.zeros(1, NDIMS, IMG_SIZE, IMG_SIZE, IMG_SIZE)
        result = om.apply_def(img=big_vol, ddf=small_ddf)
        assert list(result.shape[2:]) == [64, 64, 64]


class TestPredict:
    """Test 6: predict with random weights produces correct DDF shape.

    Uses IMG_SIZE=64 because RecMutAttnNet has 5 hierarchy levels:
    32→16→8→4→2→1 bottleneck breaks InstanceNorm (no running stats).
    """

    def test_predict_shape(self):
        sz = 64
        om = _make_omorpher(img_size=sz)
        img = torch.rand([1, 1] + [sz] * NDIMS)
        om.set_init_img(img.numpy()[0, 0])
        om.predict(T=[0, 2])
        ddf = om.get_def()
        assert ddf.shape == (1, NDIMS, sz, sz, sz)

    def test_predict_intermediate(self):
        sz = 64
        om = _make_omorpher(img_size=sz)
        img = torch.rand([1, 1] + [sz] * NDIMS)
        om.set_init_img(img.numpy()[0, 0])
        om.predict(T=[0, 4], t_save=[3, 1])
        intermediates = om.get_def(t_list=[3, 1])
        assert isinstance(intermediates, dict)

    def test_chaining(self):
        sz = 64
        om = _make_omorpher(img_size=sz)
        img = torch.rand([1, 1] + [sz] * NDIMS)
        result = om.set_init_img(img.numpy()[0, 0]).predict(T=[0, 2])
        assert result is om


class TestFinetune:
    """Test 7: finetune_step with dummy data.

    Uses IMG_SIZE=64 because at 32 the bottleneck hits 1x1x1 and
    InstanceNorm fails in training mode.
    """

    def test_finetune_roundtrip(self):
        ft_size = 64
        om = _make_omorpher(img_size=ft_size, batchsize=1)
        om.finetune_setup(lr=1e-3)
        vol = _rand_vol(S=ft_size)
        losses = om.finetune_step(vol)
        assert "loss_total" in losses
        assert "loss_grad" in losses
        assert isinstance(losses["loss_total"], float)
        om.finetune_teardown()


class TestSetters:
    """Test input setters."""

    def test_set_init_def_random(self):
        om = _make_omorpher()
        om.set_init_img(_rand_vol().numpy()[0, 0])
        om.set_init_def(None)  # should generate random
        assert om._init_ddf is not None
        assert not torch.all(om._init_ddf == 0)

    def test_set_init_def_provided(self):
        om = _make_omorpher()
        om.set_init_img(_rand_vol().numpy()[0, 0])
        custom_ddf = np.zeros([1, NDIMS] + [IMG_SIZE] * NDIMS)
        om.set_init_def(custom_ddf)
        assert torch.all(om._init_ddf == 0)

    def test_set_cond_img_default(self):
        om = _make_omorpher()
        om.set_init_img(_rand_vol().numpy()[0, 0])
        om.set_cond_img(None)
        assert om._cond_img is not None
        assert om._cond_img.shape == (1, 1, IMG_SIZE, IMG_SIZE, IMG_SIZE)

    def test_set_cond_txt_numpy(self):
        om = _make_omorpher()
        emb = np.random.randn(1024).astype(np.float32)
        om.set_cond_txt(emb)
        assert om._cond_txt is not None
        assert om._cond_txt.shape == (1, 1024)

    def test_set_init_img_with_ddf(self):
        om = _make_omorpher()
        vol = np.random.rand(40, 40, 40).astype(np.float32)
        ddf = np.zeros([1, NDIMS, IMG_SIZE, IMG_SIZE, IMG_SIZE], dtype=np.float32)
        om.set_init_img((vol, ddf))
        assert om._init_img is not None
        assert om._init_ddf is not None


# ================================================================
# 2. Cross-validation / alignment tests
# ================================================================

def _build_shared_weights():
    """Build matching OMorpher + DeformDDPM with identical random weights."""
    cfg = {**BASE_CONFIG, "inf_mode": False}  # match DeformDDPM default
    Net = get_net(NET_NAME)

    network = Net(
        n_steps=TIMESTEPS, ndims=NDIMS, num_input_chn=1, res=IMG_SIZE,
    )
    ddpm = DeformDDPM(
        network=network,
        n_steps=TIMESTEPS,
        image_chw=[1] + [IMG_SIZE] * NDIMS,
        device=DEVICE,
        batch_size=1,
        img_pad_mode="zeros",
        ddf_pad_mode="border",
        padding_mode="border",
        v_scale=cfg["v_scale"],
    )
    ddpm.eval()

    om = OMorpher(config=cfg, checkpoint_path=None, device=DEVICE)
    # Copy weights from the DeformDDPM's network to OMorpher's network
    om.network.load_state_dict(ddpm.network.state_dict())
    om.network.eval()

    return om, ddpm


class TestDDFScaleAlignment:
    """Test 10: _get_ddf_scale matches DeformDDPM._get_ddf_scale."""

    def test_all_timesteps(self):
        om, ddpm = _build_shared_weights()
        for t_val in [1, 5, 10, 20, 40, 60, 80]:
            t = torch.tensor([t_val])
            r1, m1, v1 = om._get_ddf_scale(t)
            r2, m2, v2 = ddpm._get_ddf_scale(t)
            assert r1 == r2, f"rec_num mismatch at t={t_val}"
            assert torch.equal(m1, m2), f"mul_num_ddf mismatch at t={t_val}"
            assert torch.equal(v1, v2), f"mul_num_dvf mismatch at t={t_val}"


class TestRandomDDFAlignment:
    """Test 9: _get_random_ddf matches DeformDDPM._get_random_ddf with same seed.

    Uses ndims=2, IMG_SIZE=128 so that ctl_sz=32, scale_num=5,
    len(ctl_szs_all)=5 > select_num=4 — avoiding a known unbound-variable
    bug in the original DeformDDPM._random_ddf_generate at smaller sizes.
    """

    def test_same_seed(self):
        align_size = 128
        align_ndims = 2
        cfg = {**BASE_CONFIG, "img_size": align_size, "ndims": align_ndims}
        Net = get_net(NET_NAME)
        network = Net(n_steps=TIMESTEPS, ndims=align_ndims, num_input_chn=1, res=align_size)
        ddpm = DeformDDPM(
            network=network, n_steps=TIMESTEPS,
            image_chw=[1] + [align_size] * align_ndims, device=DEVICE,
            batch_size=1, img_pad_mode="zeros", ddf_pad_mode="border",
            padding_mode="border", v_scale=cfg["v_scale"],
        )
        ddpm.eval()

        om = OMorpher(config=cfg, checkpoint_path=None, device=DEVICE)
        om.network.load_state_dict(ddpm.network.state_dict())
        om.network.eval()

        img = torch.rand([1, 1] + [align_size] * align_ndims).to(DEVICE)
        t = torch.tensor([5])

        # OMorpher
        torch.manual_seed(42)
        np.random.seed(42)
        random_mod = __import__("random")
        random_mod.seed(42)
        warped_om, dvf_om, ddf_om = om._get_random_ddf(img, t)

        # DeformDDPM
        torch.manual_seed(42)
        np.random.seed(42)
        random_mod.seed(42)
        warped_ddpm, dvf_ddpm, ddf_ddpm = ddpm._get_random_ddf(img, t)

        assert torch.allclose(ddf_om, ddf_ddpm, atol=1e-5), "DDFs do not match"
        assert torch.allclose(dvf_om, dvf_ddpm, atol=1e-5), "DVFs do not match"
        assert torch.allclose(warped_om, warped_ddpm, atol=1e-5), "Warped images do not match"


class TestConditioningAlignment:
    """Test 11: _proc_cond_img matches DeformDDPM.proc_cond_img."""

    def _test_proc_type(self, proc_type):
        om, ddpm = _build_shared_weights()
        img = _rand_vol().to(DEVICE)

        torch.manual_seed(99)
        np.random.seed(99)
        random_mod = __import__("random")
        random_mod.seed(99)
        out_om, mask_om, cr_om = om._proc_cond_img(img, proc_type=proc_type)

        torch.manual_seed(99)
        np.random.seed(99)
        random_mod.seed(99)
        out_ddpm, mask_ddpm, cr_ddpm = ddpm.proc_cond_img(img, proc_type=proc_type)

        assert torch.allclose(out_om, out_ddpm, atol=1e-5), f"Proc image mismatch for {proc_type}"

    def test_uncon(self):
        self._test_proc_type("uncon")

    def test_none(self):
        self._test_proc_type("none")

    def test_adding(self):
        self._test_proc_type("adding")

    def test_independ(self):
        self._test_proc_type("independ")

    def test_slice(self):
        self._test_proc_type("slice")

    def test_downsample(self):
        self._test_proc_type("downsample")


class TestApplyDDFAlignment:
    """Test 8: _apply_ddf matches OM_reg_flexres.apply_ddf."""

    def test_vs_flexres(self):
        om = _make_omorpher()
        vol = _rand_vol().to(DEVICE)
        ddf = torch.randn(1, NDIMS, IMG_SIZE, IMG_SIZE, IMG_SIZE, device=DEVICE) * 0.01

        # OMorpher version
        out_om = om._apply_ddf(vol, ddf, padding_mode="border")

        # Inline reimplementation of OM_reg_flexres.apply_ddf for comparison
        ndims = 3
        img_sz = list(vol.shape[2:])
        max_sz = torch.reshape(
            torch.tensor(img_sz, dtype=torch.float32, device=DEVICE),
            [1, ndims] + [1] * ndims)
        ref_grid = torch.reshape(
            torch.stack(torch.meshgrid(
                [torch.arange(s, device=DEVICE) for s in img_sz], indexing='ij'), 0),
            [1, ndims] + img_sz)
        img_shape = torch.reshape(
            torch.tensor([(s - 1) / 2. for s in img_sz], dtype=torch.float32, device=DEVICE),
            [1] + [1] * ndims + [ndims])
        grid = torch.flip(
            (ddf * max_sz + ref_grid).permute(
                [0] + list(range(2, 2 + ndims)) + [1]) / img_shape - 1,
            dims=[-1])
        out_ref = F.grid_sample(vol, grid.float(), mode="bilinear",
                                padding_mode="border", align_corners=True)

        assert torch.allclose(out_om, out_ref, atol=1e-6), "apply_ddf output mismatch"


# ================================================================
# Runner
# ================================================================

def run_all():
    import traceback
    test_classes = [
        TestInstantiation,
        TestStandardization,
        TestLabelStandardization,
        TestZeroDDFRoundtrip,
        TestSizeMismatch,
        TestPredict,
        TestFinetune,
        TestSetters,
        TestDDFScaleAlignment,
        TestRandomDDFAlignment,
        TestConditioningAlignment,
        TestApplyDDFAlignment,
    ]
    passed = 0
    failed = 0
    errors = []
    for cls in test_classes:
        inst = cls()
        for name in sorted(dir(inst)):
            if not name.startswith("test"):
                continue
            full_name = f"{cls.__name__}.{name}"
            try:
                getattr(inst, name)()
                passed += 1
                print(f"  PASS  {full_name}")
            except Exception as e:
                failed += 1
                errors.append((full_name, e))
                print(f"  FAIL  {full_name}: {e}")
                traceback.print_exc()
    print(f"\n{'='*60}")
    print(f"Results: {passed} passed, {failed} failed out of {passed+failed}")
    if errors:
        print("Failures:")
        for name, e in errors:
            print(f"  - {name}: {e}")
    return failed == 0


if __name__ == "__main__":
    success = run_all()
    sys.exit(0 if success else 1)