def main(mixup=False):
prefix = "Mixup" if mixup else "Large"
run = wandb.init(
name=f"Interpretability ({prefix})",
project="ct-interpretability",
dir=DEFAULT_DATA_STORAGE,
reinit=True,
)
model = get_model(mixup)
dataset = get_miccai_2d(
"test",
transform=DEGREE[model.hparams.transform_degree]["test"],
enhanced="Boundary" in model.hparams.loss_fx,
)
class_labels = dict(zip(range(1, model._n_classes), miccai.STRUCTURES))
class_labels[0] = "Void"
step = 0
for sample in tqdm(dataset):
preproc_img, masks, _, *others = sample
normalized_inp = preproc_img.unsqueeze(0).to(device)
normalized_inp.requires_grad = True
masks = _squash_masks(masks, 10, masks.device)
if len(masks.unique()) < 6:
# Only displaying structures with atleast 5 structures (excluding background)
continue
out = model(normalized_inp)
out_max = _squash_predictions(out).unsqueeze(1)
log_samples(preproc_img, masks, out_max, class_labels, step)
def segmentation_wrapper(input):
return model(input).sum(dim=(2, 3))
layer = model.unet.model[2][1].conv.unit0.conv
lgc = LayerGradCam(segmentation_wrapper, layer)
figures = []
for structure in miccai.STRUCTURES:
idx = structures.index(structure)
gc_attr = lgc.attribute(normalized_inp, target=idx)
fig, ax = viz.visualize_image_attr(
gc_attr[0].cpu().permute(1, 2, 0).detach().numpy(),
sign="all",
use_pyplot=False,
)
ax.set_title(structure)
figures.append(wandb.Image(fig))
wandb.log({"GradCam Attributions": figures}, step=step)
step += 1
run.finish()
def layer_grad_cam(model, image, results_dir, file_name):
layer_gc = LayerGradCam(model, model.skip_4[0])
attr = layer_gc.attribute(image)
attr = LayerAttribution.interpolate(attr, (121, 145, 121))
attr = attr.squeeze()
attr = attr.detach().numpy()
attr = nib.Nifti1Image(attr, affine=np.eye(4))
nib.save(attr, results_dir + file_name + "-HM.nii")
def get_grad_cam(x, y, model, is_training=True):
''' Choose the last conv layer which only has 7x7 out of 224x224 '''
with torch.enable_grad(), \
HackGradAndOutputs(is_training=is_training) as hack:
lgc = LayerGradCam(model, model.get_grad_cam_layer())
attributions = lgc.attribute(x, target=y)
attributions = F.interpolate(attributions,
size=x.shape[-2:],
mode='bilinear')
return attributions, hack.output
def run(arch, img, target):
input = Image.open(img).convert('RGB')
input = apply_transform(input)
model = models.vgg16(pretrained=True).eval()
ig = LayerGradCam(model, model.features[28])
out = FF.softmax(model(input), dim=1)
class_idx = out.max(1)[-1].item()
attr = ig.attribute(input, target=target)
attr = LayerAttribution.interpolate(attr, (224, 224))
attr = (attr - attr.min()) / (attr.max() - attr.min())
#attr=attr.squeeze(0).squeeze(0)
#print('IG Attributions:',attr, attr.shape)
return attr
def explanation(self, dataindex):
oldIndices = self.unknown.indices.copy()
self.unknown.indices = dataindex
datasetLoader = torch.utils.data.DataLoader(dataset=self.unknown,
batch_size=1,
shuffle=False)
self.model.eval()
# for param in self.model.parameters():
# param.requires_grad = False
avg_loss = []
#Dont forget to replace indices at end ##########
layer_gc = LayerGradCam(self.model, self.model.layer1[0].conv2)
#deep lift
dl = LayerDeepLift(self.model, self.model.layer1[0].conv2)
# atrr = []
plt.figure(figsize=(18, 10))
for i, batch in enumerate(datasetLoader):
lb = batch[1].to(device)
print(len(lb))
img = batch[0].to(device)
# plt.subplot(2,1,1)
# plt.imshow(img.squeeze().cpu().numpy())
lbin = batch[1].cpu().numpy()
print(lbin)
pred = self.model(img)
predlb = torch.argmax(pred, 1)
print('Prediction label is :', predlb.cpu().numpy())
print('Ground Truth label is: ', lb.cpu().numpy())
# gc_attr = layer_gc.attribute(img, target=int(lbin[0]))
gc_attr = layer_gc.attribute(img, target=int(predlb.cpu().numpy()))
upsampled_attr = LayerAttribution.interpolate(gc_attr, (28, 28))
base = torch.zeros([1, 1, 28, 28]).to(device)
de_attr = dl.attribute(img, base, target=int(lbin[0]))
dl_upsampled_attr = LayerAttribution.interpolate(de_attr, (28, 28))
# upsampled_attr = LayerAttribution.interpolate(gc_attr, (28, 28))
# plt.subplot(2,1,2)
# plt.imshow(upsampled_attr.squeeze().detach().cpu().numpy())
# atrr.append[gc_attr]
print("done ...")
# print(gc_attr,upsampled_attr.squeeze().detach().cpu().numpy())
# plt.show()
return img, gc_attr, upsampled_attr.squeeze().detach().cpu().numpy(
), dl_upsampled_attr.squeeze().detach().cpu().numpy()
def explain_gradXact(model, node_idx, x, edge_index, target, include_edges=None):
# Captum default implementation of LayerGradCam does not average over nodes for different channels because of
# different assumptions on tensor shapes
input_mask = x.clone().requires_grad_(True).to(device)
layers = get_all_convolution_layers(model)
node_attrs = []
for layer in layers:
layer_gc = LayerGradCam(model_forward_node, layer)
node_attr = layer_gc.attribute(input_mask, target=target, additional_forward_args=(model, edge_index, node_idx))
node_attr = node_attr.cpu().detach().numpy().ravel()
node_attrs.append(node_attr)
node_attr = np.array(node_attrs).mean(axis=0)
edge_mask = node_attr_to_edge(edge_index, node_attr)
return edge_mask
def __init__(self, trainer):
CaptumDerivative.__init__(self, trainer)
model = self.trainer
adaptive_idx = 0
self.configs.update({"relu_attributions": True}) # As in original GradCam paper
found_avg_pool = False
for adaptive_idx, mod in enumerate(trainer.model.children()):
if isinstance(mod, nn.AdaptiveAvgPool2d):
found_avg_pool = True
break
assert found_avg_pool, "This implementation assumes that final spatial dimension is reduced via AvgPool." \
"If you want to use it, change the model accordingly or update this code."
layer = model.trainer.model[adaptive_idx - 1]
LayerGradCam.__init__(self, model, layer)
def validate_explanation(self):
predlist = torch.zeros(0, dtype=torch.long, device='cpu')
lbllist = torch.zeros(0, dtype=torch.long, device='cpu')
scores = []
layer_gc = LayerGradCam(self.model, self.model.layer1[0].conv2)
for i, batch in enumerate(self.test_loader):
lb = batch[1].to(device)
img = batch[0].to(device)
pred = self.model(img)
predlb = torch.argmax(pred, 1)
gc_attr = layer_gc.attribute(img,
target=predlb,
relu_attributions=False)
upsampled_attr = LayerAttribution.interpolate(gc_attr, (64, 64))
sz = upsampled_attr.size()
x = upsampled_attr.view(sz[0], sz[1], -1)
# print(x.size())
upsampled_attr_soft = F.softmax(x, dim=2).view_as(upsampled_attr)
# print(upsampled_attr_soft.size())
# upsampled_attr = F.softmax(upsampled_attr)
# gc_attr = layer_gc.attribute(img, target=lb, relu_attributions=False)
# upsampled_attrB = LayerAttribution.interpolate(gc_attr, (64, 64))
# Append batch prediction results
predlist = torch.cat(
[predlist,
upsampled_attr_soft.detach().squeeze().cpu()])
lbllist = torch.cat([
lbllist, self.sintetic[:upsampled_attr_soft.size(
)[0], :, :, :].squeeze().cpu()
],
dim=0)
print(predlist.size, lbllist.size)
final_prec, final_rec, final_corr = self.calculate_measures(
lbllist, predlist)
# return final_prec, final_rec, final_corr
# Save checkpoint
print('Final validation result...')
print("- final explanation percision: {:.3f}".format(final_prec))
print("- final explanation recal: {:.3f}".format(final_rec))
print("- final explanation correlation: {:.3f}".format(final_corr))
def cal(name, model, image_orl, size, label_true):
func = make_transform(args, mode="inference")
image = func(image_orl)
image = image.unsqueeze(0)
image = image.to(args.gpu, dtype=torch.float32)
output = model(image)
output = F.softmax(output, dim=1)
prediction_score, pred_label_idx = torch.topk(output, 1)
pred_label_idx.squeeze()
predicted_label = str(pred_label_idx.item())
if int(predicted_label) != 1 and int(predicted_label) != label_true:
# print("predict wrong")
count.append(1)
return None
# print('Predicted:', predicted_label, '(', prediction_score.squeeze().item(), ')')
if args.att_type == "GradCam":
gradients = LayerGradCam(model, layer=model.layer4)
att_map = make_grad(name, gradients, image, image_orl, 'GradCam', size)
if args.att_type == "DeepLIFT":
gradients = LayerDeepLift(model, layer=model.layer4)
att_map = make_grad(name, gradients, image, image_orl, 'DeepLIFT', size)
return att_map
def compute_gradcam(self, img_path, target):
# open image
img, transformed_img, input = self.open_image(img_path)
# grad cam
input.requires_grad = True
gradcam = LayerGradCam(self.model, self.layer)
attr = gradcam.attribute(input, target)
cam = (attr.squeeze().cpu().detach().numpy())
cam = np.maximum(cam, 0)
cam = cv2.resize(cam, input.shape[2:])
cam = cam - np.min(cam)
cam = cam / np.max(cam)
return cam
def compute_attr_one_pixel_target(x, net, spatial_coords, method, **kwargs):
# x is a single input tensor, i.e. shape (batch_size,channel_size,H,W)=(1,1,28,28)
from captum.attr import LayerGradCam, Deconvolution, GuidedBackprop
if 'target' in kwargs:
target = kwargs['target']
if 'wrapper_output' in kwargs:
output_mode = kwargs['wrapper_output']
else:
output_mode = 'yg_pixel'
idx,idy = spatial_coords
wnet = WrapperNet(net, output_mode=output_mode, spatial_coords=(idx,idy))
if method=='gradCAM':
xai = LayerGradCam(wnet, wnet.main_net.channel_adj)
elif method=='deconv':
xai = Deconvolution(wnet)
elif method=='GuidedBP':
xai = GuidedBackprop(wnet)
if method in ['gradCAM', 'deconv', 'GuidedBP']:
attr = xai.attribute(x, target=target )
elif method == 'layerAct':
attr = xai.attribute(x)
attr = attr[0][0].clone().detach().cpu().numpy()
return attr
def initialize(self, ctx): # pylint: disable=arguments-differ
"""In this initialize function, the CIFAR10 trained model is loaded and
the Integrated Gradients,occlusion and layer_gradcam Algorithm for
Captum Explanations is initialized here.
Args:
ctx (context): It is a JSON Object containing information
pertaining to the model artifacts parameters.
"""
self.manifest = ctx.manifest
properties = ctx.system_properties
model_dir = properties.get("model_dir")
print("Model dir is {}".format(model_dir))
serialized_file = self.manifest["model"]["serializedFile"]
mapping_file_path = os.path.join(model_dir, "index_to_name.json")
if os.path.exists(mapping_file_path):
with open(mapping_file_path) as fp:
self.mapping = json.load(fp)
model_pt_path = os.path.join(model_dir, serialized_file)
self.device = torch.device(
"cuda:" + str(properties.get("gpu_id")) if torch.cuda.is_available() else "cpu"
)
from cifar10_train import CIFAR10Classifier
self.model = CIFAR10Classifier()
self.model.load_state_dict(torch.load(model_pt_path))
self.model.to(self.device)
self.model.eval()
self.model.zero_grad()
logger.info("CIFAR10 model from path %s loaded successfully", model_dir)
# Read the mapping file, index to object name
mapping_file_path = os.path.join(model_dir, "class_mapping.json")
if os.path.isfile(mapping_file_path):
print("Mapping file present")
with open(mapping_file_path) as pointer:
self.mapping = json.load(pointer)
else:
print("Mapping file missing")
logger.warning("Missing the class_mapping.json file.")
self.ig = IntegratedGradients(self.model)
self.layer_gradcam = LayerGradCam(self.model, self.model.model_conv.layer4[2].conv3)
self.occlusion = Occlusion(self.model)
self.initialized = True
self.image_processing = transforms.Compose(
[
transforms.Resize(224),
transforms.CenterCrop(224),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
]
)
def initialize(self, ctx): #pylint: disable=arguments-differ
"""In this initialize function, the Titanic trained model is loaded and
the Integrated Gradients Algorithm for Captum Explanations
is initialized here.
Args:
ctx (context): It is a JSON Object containing information
pertaining to the model artifacts parameters.
"""
self.manifest = ctx.manifest
properties = ctx.system_properties
model_dir = properties.get("model_dir")
print("Model dir is {}".format(model_dir))
serialized_file = self.manifest["model"]["serializedFile"]
model_pt_path = os.path.join(model_dir, serialized_file)
self.device = torch.device( #pylint: disable=no-member
"cuda:" + str(properties.get("gpu_id"))
if torch.cuda.is_available() else "cpu")
self.model = CIFAR10CLASSIFIER()
self.model.load_state_dict(torch.load(model_pt_path))
self.model.to(self.device)
self.model.eval()
self.model.zero_grad()
logger.info("CIFAR10 model from path %s loaded successfully",
model_dir)
# Read the mapping file, index to object name
mapping_file_path = os.path.join(model_dir, "class_mapping.json")
if os.path.isfile(mapping_file_path):
print("Mapping file present")
with open(mapping_file_path) as file_pointer:
self.mapping = json.load(file_pointer)
else:
print("Mapping file missing")
logger.warning("Missing the class_mapping.json file.")
self.ig = IntegratedGradients(self.model)
self.layer_gradcam = LayerGradCam(
self.model, self.model.model_conv.layer4[2].conv3)
self.occlusion = Occlusion(self.model)
self.initialized = True
def heatmaps(args):
print('heatmaps')
PROJECT_ID = args['PROJECT_ID']
CKPT_DIR, PROJECT_DIR, MODEL_DIR, LOGGER_DIR, load_model = folder_check(PROJECT_ID, CKPT_DIR='checkpoint')
XAI_DIR = os.path.join(PROJECT_DIR,'XAI')
if not os.path.exists(XAI_DIR): os.mkdir(XAI_DIR)
from .sampler import Pytorch_GPT_MNIST_Sampler
samp = Pytorch_GPT_MNIST_Sampler(compenv_mode=None, growth_mode=None)
from .model import ResGPTNet34
net = ResGPTNet34(nG0=samp.gen.nG0, Nj=samp.gen.N_neighbor)
net = torch.load(MODEL_DIR)
net.output_mode = 'prediction_only'
net.to(device=device)
net.eval()
x, y0, yg0, ys0 = samp.get_sample_batch(class_indices=np.array(range(10)), device=device)
x.requires_grad=True
attrs = {}
SAVE_DIR = os.path.join(XAI_DIR, 'heatmaps.y0.jpeg')
from captum.attr import LayerGradCam, Deconvolution, GuidedBackprop # ShapleyValueSampling
xai = LayerGradCam(net, net.channel_adj)
attr = xai.attribute(x, target=y0).clone().detach().cpu().numpy()
attrs['gradCAM'] = attr
xai = Deconvolution(net)
attr = xai.attribute(x, target=y0).clone().detach().cpu().numpy()
attrs['deconv'] = attr
xai = GuidedBackprop(net)
attr = xai.attribute(x, target=y0).clone().detach().cpu().numpy()
attrs['GuidedBP'] = attr
arrange_heatmaps(x.clone().detach().cpu().numpy() , attrs, save_dir=SAVE_DIR)
#
# GradCAM computes the gradients of the target output with respect to the
# given layer, averages for each output channel (dimension 2 of output),
# and multiplies the average gradient for each channel by the layer
# activations. The results are summed over all channels. GradCAM is
# designed for convnets; since the activity of convolutional layers often
# maps spatially to the input, GradCAM attributions are often upsampled
# and used to mask the input.
#
# Layer attribution is set up similarly to input attribution, except that
# in addition to the model, you must specify a hidden layer within the
# model that you wish to examine. As above, when we call ``attribute()``,
# we specify the target class of interest.
#
layer_gradcam = LayerGradCam(model, model.layer3[1].conv2)
attributions_lgc = layer_gradcam.attribute(input_img, target=pred_label_idx)
_ = viz.visualize_image_attr(attributions_lgc[0].cpu().permute(
1, 2, 0).detach().numpy(),
sign="all",
title="Layer 3 Block 1 Conv 2")
##########################################################################
# We’ll use the convenience method ``interpolate()`` in the
# `LayerAttribution <https://captum.ai/api/base_classes.html?highlight=layerattribution#captum.attr.LayerAttribution>`__
# base class to upsample this attribution data for comparison to the input
# image.
#
upsamp_attr_lgc = LayerAttribution.interpolate(attributions_lgc,
def get_grad_imp(model,
X,
y=None,
mode='grad',
return_y=False,
clip=False,
baselines=None):
X.requires_grad_(True)
# X = X.cuda()
if mode in ['grad']:
logits = model(X)
if y is None:
y = logits.argmax(dim=1)
attributions = torch.autograd.grad(
logits[torch.arange(len(logits)), y].sum(), X)[0].detach()
else:
if y is None:
with torch.no_grad():
logits = model(X)
y = logits.argmax(dim=1)
if mode == 'deeplift':
dl = DeepLift(model)
attributions = dl.attribute(inputs=X, baselines=0., target=y)
attributions = attributions.detach()
# attributions = (attributions.detach() ** 2).sum(dim=1, keepdim=True)
# elif mode in ['deepliftshap', 'deepliftshap_mean']:
elif mode in ['deepliftshap']:
dl = DeepLiftShap(model)
attributions = []
for idx in range(0, len(X), 2):
the_x, the_y = X[idx:(idx + 2)], y[idx:(idx + 2)]
attribution = dl.attribute(inputs=the_x,
baselines=baselines,
target=the_y)
attributions.append(attribution.detach())
attributions = torch.cat(attributions, dim=0)
# attributions = dl.attribute(inputs=X, baselines=baselines, target=y).detach()
# if mode == 'deepliftshap':
# attributions = (attributions ** 2).sum(dim=1, keepdim=True)
# else:
# attributions = (attributions).mean(dim=1, keepdim=True)
elif mode in ['gradcam']:
orig_lgc = LayerGradCam(model, model.body[0])
attributions = orig_lgc.attribute(X, target=y)
attributions = F.interpolate(attributions,
size=X.shape[-2:],
mode='bilinear')
else:
raise NotImplementedError(f'${mode} is not specified.')
# Do clipping!
if clip:
attributions = myclip(attributions)
X.requires_grad_(False)
if not return_y:
return attributions
return attributions, y
def get_attribution(real_img,
fake_img,
real_class,
fake_class,
net_module,
checkpoint_path,
input_shape,
channels,
methods=["ig", "grads", "gc", "ggc", "dl", "ingrad", "random", "residual"],
output_classes=6,
downsample_factors=[(2,2), (2,2), (2,2), (2,2)]):
imgs = [image_to_tensor(normalize_image(real_img).astype(np.float32)),
image_to_tensor(normalize_image(fake_img).astype(np.float32))]
classes = [real_class, fake_class]
net = init_network(checkpoint_path, input_shape, net_module, channels, output_classes=output_classes,eval_net=True, require_grad=False,
downsample_factors=downsample_factors)
attrs = []
attrs_names = []
if "residual" in methods:
res = np.abs(real_img - fake_img)
res = res - np.min(res)
attrs.append(torch.tensor(res/np.max(res)))
attrs_names.append("residual")
if "random" in methods:
rand = np.abs(np.random.randn(*np.shape(real_img)))
rand = np.abs(scipy.ndimage.filters.gaussian_filter(rand, 4))
rand = rand - np.min(rand)
rand = rand/np.max(np.abs(rand))
attrs.append(torch.tensor(rand))
attrs_names.append("random")
if "gc" in methods:
net.zero_grad()
last_conv_layer = [(name,module) for name, module in net.named_modules() if type(module) == torch.nn.Conv2d][-1]
layer_name = last_conv_layer[0]
layer = last_conv_layer[1]
layer_gc = LayerGradCam(net, layer)
gc_real = layer_gc.attribute(imgs[0], target=classes[0])
gc_fake = layer_gc.attribute(imgs[1], target=classes[1])
gc_real = project_layer_activations_to_input_rescale(gc_real.cpu().detach().numpy(), (input_shape[0], input_shape[1]))
gc_fake = project_layer_activations_to_input_rescale(gc_fake.cpu().detach().numpy(), (input_shape[0], input_shape[1]))
attrs.append(torch.tensor(gc_real[0,0,:,:]))
attrs_names.append("gc_real")
attrs.append(torch.tensor(gc_fake[0,0,:,:]))
attrs_names.append("gc_fake")
# SCAM
gc_diff_0, gc_diff_1 = get_sgc(real_img, fake_img, real_class,
fake_class, net_module, checkpoint_path,
input_shape, channels, None, output_classes=output_classes,
downsample_factors=downsample_factors)
attrs.append(gc_diff_0)
attrs_names.append("gc_diff_0")
attrs.append(gc_diff_1)
attrs_names.append("gc_diff_1")
if "ggc" in methods:
net.zero_grad()
last_conv = [module for module in net.modules() if type(module) == torch.nn.Conv2d][-1]
guided_gc = GuidedGradCam(net, last_conv)
ggc_real = guided_gc.attribute(imgs[0], target=classes[0])
ggc_fake = guided_gc.attribute(imgs[1], target=classes[1])
attrs.append(ggc_real[0,0,:,:])
attrs_names.append("ggc_real")
attrs.append(ggc_fake[0,0,:,:])
attrs_names.append("ggc_fake")
net.zero_grad()
gbp = GuidedBackprop(net)
gbp_real = gbp.attribute(imgs[0], target=classes[0])
gbp_fake = gbp.attribute(imgs[1], target=classes[1])
attrs.append(gbp_real[0,0,:,:])
attrs_names.append("gbp_real")
attrs.append(gbp_fake[0,0,:,:])
attrs_names.append("gbp_fake")
ggc_diff_0 = gbp_real[0,0,:,:] * gc_diff_0
ggc_diff_1 = gbp_fake[0,0,:,:] * gc_diff_1
attrs.append(ggc_diff_0)
attrs_names.append("ggc_diff_0")
attrs.append(ggc_diff_1)
attrs_names.append("ggc_diff_1")
# IG
if "ig" in methods:
baseline = image_to_tensor(np.zeros(input_shape, dtype=np.float32))
net.zero_grad()
ig = IntegratedGradients(net)
ig_real, delta_real = ig.attribute(imgs[0], baseline, target=classes[0], return_convergence_delta=True)
ig_fake, delta_fake = ig.attribute(imgs[1], baseline, target=classes[1], return_convergence_delta=True)
ig_diff_0, delta_diff = ig.attribute(imgs[0], imgs[1], target=classes[0], return_convergence_delta=True)
ig_diff_1, delta_diff = ig.attribute(imgs[1], imgs[0], target=classes[1], return_convergence_delta=True)
attrs.append(ig_real[0,0,:,:])
attrs_names.append("ig_real")
attrs.append(ig_fake[0,0,:,:])
attrs_names.append("ig_fake")
attrs.append(ig_diff_0[0,0,:,:])
attrs_names.append("ig_diff_0")
attrs.append(ig_diff_1[0,0,:,:])
attrs_names.append("ig_diff_1")
# DL
if "dl" in methods:
net.zero_grad()
dl = DeepLift(net)
dl_real = dl.attribute(imgs[0], target=classes[0])
dl_fake = dl.attribute(imgs[1], target=classes[1])
dl_diff_0 = dl.attribute(imgs[0], baselines=imgs[1], target=classes[0])
dl_diff_1 = dl.attribute(imgs[1], baselines=imgs[0], target=classes[1])
attrs.append(dl_real[0,0,:,:])
attrs_names.append("dl_real")
attrs.append(dl_fake[0,0,:,:])
attrs_names.append("dl_fake")
attrs.append(dl_diff_0[0,0,:,:])
attrs_names.append("dl_diff_0")
attrs.append(dl_diff_1[0,0,:,:])
attrs_names.append("dl_diff_1")
# INGRAD
if "ingrad" in methods:
net.zero_grad()
saliency = Saliency(net)
grads_real = saliency.attribute(imgs[0],
target=classes[0])
grads_fake = saliency.attribute(imgs[1],
target=classes[1])
attrs.append(grads_real[0,0,:,:])
attrs_names.append("grads_real")
attrs.append(grads_fake[0,0,:,:])
attrs_names.append("grads_fake")
net.zero_grad()
input_x_gradient = InputXGradient(net)
ingrad_real = input_x_gradient.attribute(imgs[0], target=classes[0])
ingrad_fake = input_x_gradient.attribute(imgs[1], target=classes[1])
ingrad_diff_0 = grads_fake * (imgs[0] - imgs[1])
ingrad_diff_1 = grads_real * (imgs[1] - imgs[0])
attrs.append(torch.abs(ingrad_real[0,0,:,:]))
attrs_names.append("ingrad_real")
attrs.append(torch.abs(ingrad_fake[0,0,:,:]))
attrs_names.append("ingrad_fake")
attrs.append(torch.abs(ingrad_diff_0[0,0,:,:]))
attrs_names.append("ingrad_diff_0")
attrs.append(torch.abs(ingrad_diff_1[0,0,:,:]))
attrs_names.append("ingrad_diff_1")
attrs = [a.detach().cpu().numpy() for a in attrs]
attrs_norm = [a/np.max(np.abs(a)) for a in attrs]
return attrs_norm, attrs_names
attr_gbp, i = torch.max(attr_gbp, dim=1)
attr_gbp = attr_gbp.unsqueeze(0)
visualize_attr_maps('visualization/Captum_Guided_BackProp.png', X, y,
class_names, attr_gbp, ['GuidedBackprop'],
lambda attr: attr.detach().numpy())
# Try out different layers
# and see observe how the attributions change
layer = model.features[3]
layer_act = LayerActivation(model, layer)
layer_act_attr = compute_attributions(layer_act, X_tensor)
layer_act_attr_sum = layer_act_attr.mean(axis=1, keepdim=True)
# Layer gradcam aggregates across all channels
layer_gradcam = LayerGradCam(model, layer)
layer_gradcam_attr = compute_attributions(layer_gradcam,
X_tensor,
target=y_tensor,
relu_attributions=True)
layer_gradcam_attr_sum = layer_gradcam_attr.mean(axis=1, keepdim=True)
layer_gradcam_attr_sum = layer_gradcam_attr_sum.permute(1, 0, 2, 3)
visualize_attr_maps('visualization/layer_gradcam.png', X, y, class_names,
layer_gradcam_attr_sum, ['layer_gradcam'],
lambda attr: attr.detach().numpy())
layer_conduct = LayerConductance(model, layer)
layer_conduct_attr = compute_attributions(layer_conduct,
X_tensor,
target=y_tensor)
layer_conduct_attr_sum = layer_conduct_attr.mean(axis=1, keepdim=True)
def train_single_scale(D, G, reals, generators, noise_maps,
input_from_prev_scale, noise_amplitudes, opt):
""" Train one scale. D and G are the current discriminator and generator, reals are the scaled versions of the
original level, generators and noise_maps contain information from previous scales and will receive information in
this scale, input_from_previous_scale holds the noise map and images from the previous scale, noise_amplitudes hold
the amplitudes for the noise in all the scales. opt is a namespace that holds all necessary parameters. """
current_scale = len(generators)
real = reals[current_scale]
keepSky = False
kernel_dims = (2, 2)
# Initialize real detector
real0 = preprocess(opt, real, keepSky)
N, C, H, W = real0.shape
scale = opt.scales[current_scale] if current_scale < len(opt.scales) else 1
if opt.cgan:
detector = PCA_Detector(opt, 'real', real0, kernel_dims)
real_detection_map = detector(real0)
detection_scale = 0.1
real_detection_map *= detection_scale
real1 = torch.cat(
[real, F.interpolate(real_detection_map, (H, W))], dim=1)
divergences = []
else:
real1 = real
if opt.game == 'mario':
token_group = MARIO_TOKEN_GROUPS
else: # if opt.game == 'mariokart':
token_group = MARIOKART_TOKEN_GROUPS
nzx = real.shape[2] # Noise size x
nzy = real.shape[3] # Noise size y
padsize = int(
1 * opt.num_layer
) # As kernel size is always 3 currently, padsize goes up by one per layer
if not opt.pad_with_noise:
pad_noise = nn.ZeroPad2d(padsize)
pad_image = nn.ZeroPad2d(padsize)
else:
pad_noise = nn.ReflectionPad2d(padsize)
pad_image = nn.ReflectionPad2d(padsize)
# setup optimizer
optimizerD = optim.Adam(D.parameters(),
lr=opt.lr_d,
betas=(opt.beta1, 0.999))
optimizerG = optim.Adam(G.parameters(),
lr=opt.lr_g,
betas=(opt.beta1, 0.999))
schedulerD = torch.optim.lr_scheduler.MultiStepLR(optimizer=optimizerD,
milestones=[1600, 2500],
gamma=opt.gamma)
schedulerG = torch.optim.lr_scheduler.MultiStepLR(optimizer=optimizerG,
milestones=[1600, 2500],
gamma=opt.gamma)
if current_scale == 0: # Generate new noise
z_opt = generate_spatial_noise([1, opt.nc_current, nzx, nzy],
device=opt.device)
z_opt = pad_noise(z_opt)
else: # Add noise to previous output
z_opt = torch.zeros([1, opt.nc_current, nzx, nzy]).to(opt.device)
z_opt = pad_noise(z_opt)
logger.info("Training at scale {}", current_scale)
for epoch in tqdm(range(opt.niter)):
step = current_scale * opt.niter + epoch
noise_ = generate_spatial_noise([1, opt.nc_current, nzx, nzy],
device=opt.device)
noise_ = pad_noise(noise_)
############################
# (1) Update D network: maximize D(x) + D(G(z))
###########################
for j in range(opt.Dsteps):
# train with real
D.zero_grad()
output = D(real1).to(opt.device)
errD_real = -output.mean()
errD_real.backward(retain_graph=True)
# train with fake
if (j == 0) & (epoch == 0):
if current_scale == 0: # If we are in the lowest scale, noise is generated from scratch
prev = torch.zeros(1, opt.nc_current, nzx,
nzy).to(opt.device)
input_from_prev_scale = prev
prev = pad_image(prev)
z_prev = torch.zeros(1, opt.nc_current, nzx,
nzy).to(opt.device)
z_prev = pad_noise(z_prev)
opt.noise_amp = 1
else: # First step in NOT the lowest scale
# We need to adapt our inputs from the previous scale and add noise to it
prev = draw_concat(generators, noise_maps, reals,
noise_amplitudes, input_from_prev_scale,
"rand", pad_noise, pad_image, opt)
# For the seeding experiment, we need to transform from token_groups to the actual token
if current_scale == (opt.token_insert + 1):
prev = group_to_token(prev, opt.token_list,
token_group)
prev = interpolate(prev,
real1.shape[-2:],
mode="bilinear",
align_corners=False)
prev = pad_image(prev)
z_prev = draw_concat(generators, noise_maps, reals,
noise_amplitudes,
input_from_prev_scale, "rec",
pad_noise, pad_image, opt)
# For the seeding experiment, we need to transform from token_groups to the actual token
if current_scale == (opt.token_insert + 1):
z_prev = group_to_token(z_prev, opt.token_list,
token_group)
z_prev = interpolate(z_prev,
real1.shape[-2:],
mode="bilinear",
align_corners=False)
opt.noise_amp = update_noise_amplitude(
z_prev, real1[:, :-1], opt)
z_prev = pad_image(z_prev)
else: # Any other step
prev = draw_concat(generators, noise_maps, reals,
noise_amplitudes, input_from_prev_scale,
"rand", pad_noise, pad_image, opt)
# For the seeding experiment, we need to transform from token_groups to the actual token
if current_scale == (opt.token_insert + 1):
prev = group_to_token(prev, opt.token_list, token_group)
prev = interpolate(prev,
real1.shape[-2:],
mode="bilinear",
align_corners=False)
prev = pad_image(prev)
# After creating our correct noise input, we feed it to the generator:
noise = opt.noise_amp * noise_ + prev
fake = G(noise.detach(),
prev,
temperature=1 if current_scale != opt.token_insert else 1)
fake0 = preprocess(opt, fake, keepSky)
if opt.cgan:
Nf, Cf, Hf, Wf = fake0.shape
fake_detection_map = detector(fake0) * detection_scale
fake1 = torch.cat(
[fake, F.interpolate(fake_detection_map, (Hf, Wf))], dim=1)
else:
fake1 = fake
# Then run the result through the discriminator
output = D(fake1.detach())
errD_fake = output.mean()
# Backpropagation
errD_fake.backward(retain_graph=False)
# Gradient Penalty
gradient_penalty = calc_gradient_penalty(D, real1, fake1,
opt.lambda_grad,
opt.device)
gradient_penalty.backward(retain_graph=False)
# Logging:
if step % 10 == 0:
wandb.log(
{
f"D(G(z))@{current_scale}":
errD_fake.item(),
f"D(x)@{current_scale}":
-errD_real.item(),
f"gradient_penalty@{current_scale}":
gradient_penalty.item()
},
step=step,
sync=False)
optimizerD.step()
############################
# (2) Update G network: maximize D(G(z))
###########################
for j in range(opt.Gsteps):
G.zero_grad()
fake = G(noise.detach(),
prev.detach(),
temperature=1 if current_scale != opt.token_insert else 1)
fake0 = preprocess(opt, fake, keepSky)
Nf, Cf, Hf, Wf = fake0.shape
if opt.cgan:
fake_detection_map = detector(fake0) * detection_scale
fake1 = torch.cat(
[fake, F.interpolate(fake_detection_map, (Hf, Wf))], dim=1)
else:
fake1 = fake
output = D(fake1)
errG = -output.mean()
errG.backward(retain_graph=False)
if opt.alpha != 0: # i. e. we are trying to find an exact recreation of our input in the lat space
Z_opt = opt.noise_amp * z_opt + z_prev
G_rec = G(
Z_opt.detach(),
z_prev,
temperature=1 if current_scale != opt.token_insert else 1)
rec_loss = opt.alpha * F.mse_loss(G_rec, real)
if opt.cgan:
div = divergence(real_detection_map,
preprocess(opt, G_rec, keepSky))
rec_loss += div
rec_loss.backward(
retain_graph=False
) # TODO: Check for unexpected argument retain_graph=True
rec_loss = rec_loss.detach()
else: # We are not trying to find an exact recreation
rec_loss = torch.zeros([])
Z_opt = z_opt
optimizerG.step()
# More Logging:
div = divergence(real_detection_map, preprocess(opt, fake, keepSky))
divergences.append(div)
# logger.info("divergence(fake) = {}", div)
if step % 10 == 0:
wandb.log(
{
f"noise_amplitude@{current_scale}": opt.noise_amp,
f"rec_loss@{current_scale}": rec_loss.item()
},
step=step,
sync=False,
commit=True)
# Rendering and logging images of levels
if epoch % 500 == 0 or epoch == (opt.niter - 1):
if opt.token_insert >= 0 and opt.nc_current == len(token_group):
token_list = [list(group.keys())[0] for group in token_group]
else:
token_list = opt.token_list
img = opt.ImgGen.render(
one_hot_to_ascii_level(fake1[:, :-1].detach(), token_list))
img2 = opt.ImgGen.render(
one_hot_to_ascii_level(
G(Z_opt.detach(),
z_prev,
temperature=1 if current_scale != opt.token_insert else
1).detach(), token_list))
real_scaled = one_hot_to_ascii_level(real1[:, :-1].detach(),
token_list)
img3 = opt.ImgGen.render(real_scaled)
wandb.log(
{
f"G(z)@{current_scale}": wandb.Image(img),
f"G(z_opt)@{current_scale}": wandb.Image(img2),
f"real@{current_scale}": wandb.Image(img3)
},
sync=False,
commit=False)
real_scaled_path = os.path.join(wandb.run.dir,
f"real@{current_scale}.txt")
with open(real_scaled_path, "w") as f:
f.writelines(real_scaled)
wandb.save(real_scaled_path)
# Learning Rate scheduler step
schedulerD.step()
schedulerG.step()
if opt.cgan:
div = divergence(real_detection_map, preprocess(opt, z_opt, keepSky))
divergences.append(div)
# visualization config
folder_name = 'gradcam'
level_name = opt.input_name.rsplit(".", 1)[0].split("_", 1)[1]
# GradCAM on D
camD = LayerGradCam(D, D.tail)
real0 = one_hot_to_ascii_level(real, opt.token_list)
real0 = opt.ImgGen.render(real0)
real0 = np.array(real0)
attr = camD.attribute(real1, target=(0, 0, 0), relu_attributions=True)
attr = LayerAttribution.interpolate(attr, (real0.shape[0], real0.shape[1]),
'bilinear')
attr = attr.permute(2, 3, 1, 0).squeeze(3)
attr = attr.detach().cpu().numpy()
fig, ax = plt.subplots(1, 1)
fig.figsize = (10, 1)
ax.imshow(rgb2gray(real0), cmap='gray', vmin=0, vmax=1)
im = ax.imshow(attr, cmap='jet', alpha=0.5)
ax.axis('off')
fig.colorbar(im, ax=ax, location='bottom', shrink=0.85)
plt.suptitle(f'cGAN {level_name} D(x)@{current_scale} ({step})')
plt.savefig(rf'{folder_name}\{level_name}_D_{current_scale}_{step}.png',
bbox_inches='tight',
pad_inches=0.1)
# plt.show()
plt.close()
# GradCAM on G
token_names = {
'M': 'Mario start',
'F': 'Mario finish',
'y': 'spiky',
'Y': 'winged spiky',
'k': 'green koopa',
'K': 'winged green koopa',
'!': 'coin [?]',
'#': 'pyramid',
'-': 'sky',
'1': 'invis. 1 up',
'2': 'invis. coin',
'L': '1 up',
'?': 'special [?]',
'@': 'special [?]',
'Q': 'coin [?]',
'!': 'coin [?]',
'C': 'coin brick',
'S': 'normal brick',
'U': 'mushroom brick',
'X': 'ground',
'E': 'goomba',
'g': 'goomba',
'k': 'green koopa',
'%': 'platform',
'|': 'platform bg',
'r': 'red koopa',
'R': 'winged red koopa',
'o': 'coin',
't': 'pipe',
'T': 'plant pipe',
'*': 'bullet bill',
'<': 'pipe top left',
'>': 'pipe top right',
'[': 'pipe left',
']': 'pipe right',
'B': 'bullet bill head',
'b': 'bullet bill body',
'D': 'used block',
}
def wrappedG(z):
return G(z, z_opt)
camG = LayerGradCam(wrappedG, G.tail[0])
z_cam = generate_spatial_noise([1, opt.nc_current, nzx, nzy],
device=opt.device)
z_cam = pad_noise(z_cam)
attrs = []
for i in range(opt.nc_current):
attr = camG.attribute(z_cam, target=(i, 0, 0), relu_attributions=True)
attr = LayerAttribution.interpolate(attr,
(real0.shape[0], real0.shape[1]),
'bilinear')
attr = attr.permute(2, 3, 1, 0).squeeze(3)
attr = attr.detach().cpu().numpy()
attrs.append(attr)
fig, axs = plt.subplots(opt.nc_current, 1)
fig.figsize = (10, opt.nc_current)
for i in range(opt.nc_current):
axs[i].axis('off')
axs[i].text(-0.1,
0.5,
token_names[opt.token_list[i]],
rotation=0,
verticalalignment='center',
horizontalalignment='right',
transform=axs[i].transAxes)
axs[i].imshow(rgb2gray(real0), cmap='gray', vmin=0, vmax=1)
im = axs[i].imshow(attrs[i], cmap='jet', alpha=0.5)
fig.colorbar(im, ax=axs, shrink=0.85)
plt.suptitle(f'cGAN {level_name} G(z)@{current_scale} ({step})')
plt.savefig(rf'{folder_name}\{level_name}_G_{current_scale}_{step}.png',
bbox_inches='tight',
pad_inches=0.1)
# plt.show()
plt.close()
# Save networks
torch.save(z_opt, "%s/z_opt.pth" % opt.outf)
save_networks(G, D, z_opt, opt)
wandb.save(opt.outf)
return z_opt, input_from_prev_scale, G, divergences
def LayerGradCAM(classifier_model,
config,
dataset_features,
GNNgraph_list,
current_fold=None,
cuda=0):
'''
Attribute to input layer using soft assign
:param classifier_model: trained classifier model
:param config: parsed configuration file of config.yml
:param dataset_features: a dictionary of dataset features obtained from load_data.py
:param GNNgraph_list: a list of GNNgraphs obtained from the dataset
:param current_fold: has no use in this method
:param cuda: whether to use GPU to perform conversion to tensor
'''
# Initialise settings
config = config
interpretability_config = config["interpretability_methods"][
"LayerGradCAM"]
dataset_features = dataset_features
assign_type = interpretability_config["assign_attribution"]
# Perform grad cam on the classifier model and on a specific layer
layer_idx = interpretability_config["layer"]
if layer_idx == 0:
gc = LayerGradCam(classifier_model, classifier_model.graph_convolution)
else:
gc = LayerGradCam(classifier_model,
classifier_model.conv_modules[layer_idx - 1])
output_for_metrics_calculation = []
output_for_generating_saliency_map = {}
# Obtain attribution score for use in qualitative metrics
tmp_timing_list = []
for GNNgraph in GNNgraph_list:
output = {'graph': GNNgraph}
for _, label in dataset_features["label_dict"].items():
# Relabel all just in case, may only relabel those that need relabelling
# if performance is poor
original_label = GNNgraph.label
GNNgraph.label = label
node_feat, n2n, subg = graph_to_tensor(
[GNNgraph], dataset_features["feat_dim"],
dataset_features["edge_feat_dim"], cuda)
start_generation = perf_counter()
attribution = gc.attribute(node_feat,
additional_forward_args=(n2n, subg,
[GNNgraph]),
target=label,
relu_attributions=True)
# Attribute to the input layer using the assign method specified
reverse_assign_tensor_list = []
for i in range(1, layer_idx + 1):
assign_tensor = classifier_model.cur_assign_tensor_list[i - 1]
max_index = torch.argmax(assign_tensor, dim=1, keepdim=True)
if assign_type == "hard":
reverse_assign_tensor = torch.transpose(
torch.zeros(assign_tensor.size()).scatter_(1,
max_index,
value=1), 0,
1)
else:
reverse_assign_tensor = torch.transpose(
assign_tensor, 0, 1)
reverse_assign_tensor_list.append(reverse_assign_tensor)
attribution = torch.transpose(attribution, 0, 1)
for reverse_tensor in reversed(reverse_assign_tensor_list):
attribution = attribution @ reverse_tensor
attribution = torch.transpose(attribution, 0, 1)
tmp_timing_list.append(perf_counter() - start_generation)
attribution_score = torch.sum(attribution, dim=1).tolist()
attribution_score = standardize_scores(attribution_score)
GNNgraph.label = original_label
output[label] = attribution_score
output_for_metrics_calculation.append(output)
execution_time = sum(tmp_timing_list) / (len(tmp_timing_list))
# Obtain attribution score for use in generating saliency map for comparison with zero tensors
if interpretability_config["sample_ids"] is not None:
if ',' in str(interpretability_config["sample_ids"]):
sample_graph_id_list = list(
map(int, interpretability_config["sample_ids"].split(',')))
else:
sample_graph_id_list = [int(interpretability_config["sample_ids"])]
output_for_generating_saliency_map.update({
"layergradcam_%s_%s" % (str(assign_type), str(label)): []
for _, label in dataset_features["label_dict"].items()
})
for index in range(len(output_for_metrics_calculation)):
tmp_output = output_for_metrics_calculation[index]
tmp_label = tmp_output['graph'].label
if tmp_output['graph'].graph_id in sample_graph_id_list:
element_name = "layergradcam_%s_%s" % (str(assign_type),
str(tmp_label))
output_for_generating_saliency_map[element_name].append(
(tmp_output['graph'], tmp_output[tmp_label]))
elif interpretability_config["number_of_samples"] > 0:
# Randomly sample from existing list:
graph_idxes = list(range(len(output_for_metrics_calculation)))
random.shuffle(graph_idxes)
output_for_generating_saliency_map.update({
"layergradcam_%s_%s" % (str(assign_type), str(label)): []
for _, label in dataset_features["label_dict"].items()
})
# Begin appending found samples
for index in graph_idxes:
tmp_label = output_for_metrics_calculation[index]['graph'].label
element_name = "layergradcam_%s_%s" % (str(assign_type),
str(tmp_label))
if len(output_for_generating_saliency_map[element_name]
) < interpretability_config["number_of_samples"]:
output_for_generating_saliency_map[element_name].append(
(output_for_metrics_calculation[index]['graph'],
output_for_metrics_calculation[index][tmp_label]))
return output_for_metrics_calculation, output_for_generating_saliency_map, execution_time
process_sequence(gradcam, test_loader, model_type, device, output_file)
#%% captum grad cam
experiment_num = 4
#test_list_full = [4,11,18,22,23,24,25,26,27,28,29,32,33,34,36,37,38,39]
test_list_full = [32, 34]
train_list = [1, 3, 5, 7, 8, 10, 12, 14, 15, 17, 19, 21]
for test in test_list_full:
test_list = [test]
loader_dict, loader_sizes = dat.init_dataset(train_list, train_list,
test_list, model_type,
config_dict)
test_loader = loader_dict['test']
captum_gc = LayerGradCam(model, target_layer)
output_file = file_dir + '/heatmaps/imageset_captum_new_' + model_type + "_" + str(
test)
process_sequence_captum(captum_gc, test_loader, model_type, device,
output_file)
#%%
finalconv_name = 'layer4'
target_layer = model.cnn._modules.get(finalconv_name)
inputs, inputs_aug, _ = next(iter(loader_dict['train']))
captum_gc = LayerGradCam(model, target_layer)
attributions = captum_gc.attribute((inputs.to(
device, dtype=torch.float), inputs_aug.to(device, dtype=torch.float)),
target=0,
relu_attributions=True)
upsampled_attr = captum_gc.interpolate(attributions, (224, 224), 'bicubic')
32,33,
34,36,
37,38,39,
45,46,47]
condition_list = ['center','right','left',
'right_less','right_less',
'right_more','right_more',
'left_less','left_less',
'left_more','left_more',
'new_tool','new_tool',
'new_material','new_material',
'center','right','left',
'z_mid','z_high','z_low']
'''
test_list_full = [34]
train_list = [1, 3, 5, 7, 8, 10, 12, 14, 15, 17, 19, 21, 41, 42]
for test in test_list_full:
test_list = [test]
loader_dict, loader_sizes = dat.init_dataset(train_list, train_list,
test_list, model_type,
config_dict)
test_loader = loader_dict['test']
captum_gc = LayerGradCam(model, target_layer)
output_file = 'heatmaps/imageset_captum' + model_type + "_" + str(test)
process_sequence_captum(captum_gc, test_loader, model_type, device,
output_file)
model.to(DEVICE)
img, img_t = load_preprocess_image(args.img, gray=args.gray)
img_t = img_t.to(DEVICE)
img_t.requires_grad = True
input_shape = img.shape[:2]
cent_idx = (img_t.shape[0] // 2, img_t.shape[1] // 2)
# Grab attribution mapping
attr_map_ = {}
for name, layer in model.named_children():
if "up" in name:
attr_map_[name] = LayerGradCam(model.forward, model.up3)
data_attr_ = {}
for name, act in attr_map_.items():
data_attr_[name] = act.attribute(img_t, target=(1, ) + cent_idx)
## Plot attributions
fig: plt.Figure = plt.figure(figsize=(12, 4))
for k, (layer_name, attrib) in enumerate(data_attr_.items()):
plt.subplot(1, 3, k + 1)
attr_arr = attrib.data.cpu().numpy()[0, 0]
plt.imshow(attr_arr)
plt.title("GradCam center px %s" % layer_name)
fig.tight_layout()
plt.show()
def vis_explanation(self, number):
if len(self.explainVis) == 0:
for i, batch in enumerate(self.test_loader):
self.explainVis = batch
break
# oldIndices = self.test_loader.indices.copy()
# self.test_loader.indices = self.test_loader.indices[:2]
# datasetLoader = self.test_loader
layer_gc = LayerGradCam(self.model, self.model.layer2[1].conv2)
# for i, batch in enumerate(datasetLoader):
lb = self.explainVis[1].to(device)
# print(len(lb))
img = self.explainVis[0].to(device)
# plt.subplot(2,1,1)
# plt.imshow(img.squeeze().cpu().numpy())
pred = self.model(img)
predlb = torch.argmax(pred, 1)
imgCQ = img.clone()
# print('Prediction label is :',predlb.cpu().numpy())
# print('Ground Truth label is: ',lb.cpu().numpy())
##explain to me :
gc_attr = layer_gc.attribute(imgCQ,
target=predlb,
relu_attributions=False)
upsampled_attr = LayerAttribution.interpolate(gc_attr, (64, 64))
gc_attr = layer_gc.attribute(imgCQ, target=lb, relu_attributions=False)
upsampled_attrB = LayerAttribution.interpolate(gc_attr, (64, 64))
if not os.path.exists('./pic'):
os.mkdir('./pic')
####PLot################################################
plotMe = viz.visualize_image_attr(
upsampled_attr[7].detach().cpu().numpy().transpose([1, 2, 0]),
original_image=img[7].detach().cpu().numpy().transpose([1, 2, 0]),
method='heat_map',
sign='all',
plt_fig_axis=None,
outlier_perc=2,
cmap='inferno',
alpha_overlay=0.2,
show_colorbar=True,
title=str(predlb[7]),
fig_size=(8, 10),
use_pyplot=True)
plotMe[0].savefig('./pic/' + str(number) + 'NotEQPred.jpg')
################################################
plotMe = viz.visualize_image_attr(
upsampled_attrB[7].detach().cpu().numpy().transpose([1, 2, 0]),
original_image=img[7].detach().cpu().numpy().transpose([1, 2, 0]),
method='heat_map',
sign='all',
plt_fig_axis=None,
outlier_perc=2,
cmap='inferno',
alpha_overlay=0.9,
show_colorbar=True,
title=str(lb[7].cpu()),
fig_size=(8, 10),
use_pyplot=True)
plotMe[0].savefig('./pic/' + str(number) + 'NotEQLabel.jpg')
################################################
outImg = img[7].squeeze().detach().cpu().numpy()
fig2 = plt.figure(figsize=(12, 12))
prImg = plt.imshow(outImg)
fig2.savefig('./pic/' + str(number) + 'NotEQOrig.jpg')
################################################
fig = plt.figure(figsize=(15, 10))
ax = fig.add_subplot(111, projection='3d')
z = upsampled_attr[7].squeeze().detach().cpu().numpy()
x = np.arange(0, 64, 1)
y = np.arange(0, 64, 1)
X, Y = np.meshgrid(x, y)
plll = ax.plot_surface(X, Y, z, cmap=cm.coolwarm)
# Customize the z axis.
# ax.set_zlim(np.min(z)+0.1*np.min(z),np.max(z)+0.1*np.max(z))
ax.set_zlim(-0.02, 0.1)
ax.zaxis.set_major_locator(LinearLocator(10))
ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f'))
# Add a color bar which maps values to colors.
fig.colorbar(plll, shrink=0.5, aspect=5)
fig.savefig('./pic/' + str(number) + 'NotEQ3D.jpg')
def train_known_expl(self, n_epochs, ite, max_ite):
self.model.train()
avg_loss = []
# if os.path.exists('./checkpoint'):
# try:
# print('model found ...')
# self.model.load_state_dict(torch.load('./checkpoint/resnet18.pth'))
# print('Model loaded sucessfully.')
# continue
# except:
# print('Not found any model ... ')
optim = torch.optim.Adam(self.model.parameters(),
lr=0.001,
weight_decay=0)
# lossOH = OhemCELoss(0.2,50) #cause batch is 100
criteria1 = nn.CrossEntropyLoss()
criteria2 = nn.BCELoss()
criteria21 = nn.BCEWithLogitsLoss()
criteriaL2 = nn.MSELoss()
layer_gc = LayerGradCam(self.model, self.model.layer2[1].conv2)
# layer_DLS = LayerDeepLiftShap(self.model, self.model.layer1[0].conv2, multiply_by_inputs=False)
looss = nn.CrossEntropyLoss()
print('Init known training with explanation...')
number = 0
# wandb.watch(self.model, log='all')
for epoch in range(n_epochs):
for i, batch in enumerate(self.KnownLoader):
lb = batch[1].to(device)
# print(batch[0].size())
# # img = batch[0].to(device)
# img = F.interpolate(batch[0],(100,1,224,224))
# print(img.size())
img = batch[0].to(device)
# print(img.size())
#define mask
maskLb = batch[0].clone()
maskLb = maskLb.squeeze()
maskLb[maskLb == -0.5] = 0
maskLb[maskLb != 0] = 1
maskLb = maskLb.to(device)
# Training
optim.zero_grad()
# a,b,c,out = self.model(img)
out = self.model(img)
predlb = torch.argmax(out, 1)
# predlb = predlb.cpu().numpy()
# print('Prediction label is :',predlb.cpu().numpy())
# print('Ground Truth label is: ',lb.cpu().numpy())
##explain to me :
gc_attr = layer_gc.attribute(img,
target=predlb,
relu_attributions=True)
upsampled_attr = LayerAttribution.interpolate(
gc_attr, (64, 64))
gc_attr = layer_gc.attribute(img,
target=lb,
relu_attributions=True)
upsampled_attrB = LayerAttribution.interpolate(
gc_attr, (64, 64))
exitpic = upsampled_attrB.clone()
exitpic = exitpic.detach().cpu().numpy()
# wandb.log({"Examples": exitpic})
print(self.model.layer2[1].conv2.grads.data)
# upsampled_attr_B = upsampled_attr.clone()
# sz = upsampled_attr.size()
# x = upsampled_attr_B.view(sz[0],sz[1], -1)
# # print(x.size())
# upsampled_attr_B = F.softmax(x,dim=2).view_as(upsampled_attr_B)
########################################
# grid = torchvision.utils.make_grid(img)
# self.writer.add_image('images', grid, 0)
# self.writer.add_graph(self.model, img)
# baseLine = torch.zeros(img.size())
# # baseLine = baseLine[:1]
# # print(baseLine.size())
# baseLine = baseLine.to(device)
# DLS_attr,delta = layer_DLS.attribute(img,baseLine,target=predlb,return_convergence_delta =True)
# upsampled_attrDLS = LayerAttribution.interpolate(DLS_attr, (64, 64))
# upsampled_attrDLSSum = torch.sum(upsampled_attrDLS,dim=(1),keepdim=True)
# print(upsampled_attrDLSSum.size())
# print(delta.size(),DLS_attr.size())
# if number % 60 ==0:
# z = torch.eq(lb,predlb)
# z = ~z
# z = z.nonzero()
# try:
# z = z.cpu().numpy()[-1]
# except:
# z = [0]
# # if z.size().cpu()>0:
# print(lb[z[0]],predlb[z[0]],z[0])
# ################################################
# plotMe = viz.visualize_image_attr(upsampled_attr[z[0]].detach().cpu().numpy().transpose([1,2,0]),
# original_image=img[z[0]].detach().cpu().numpy().transpose([1,2,0]),
# method='heat_map',
# sign='absolute_value', plt_fig_axis=None, outlier_perc=2,
# cmap='inferno', alpha_overlay=0.2, show_colorbar=True,
# title=str(predlb[z[0]]),
# fig_size=(8, 10), use_pyplot=True)
# plotMe[0].savefig(str(number)+'NotEQPred.jpg')
# ################################################
# plotMe = viz.visualize_image_attr(upsampled_attrB[z[0]].detach().cpu().numpy().transpose([1,2,0]),
# original_image=img[z[0]].detach().cpu().numpy().transpose([1,2,0]),
# method='heat_map',
# sign='absolute_value', plt_fig_axis=None, outlier_perc=2,
# cmap='inferno', alpha_overlay=0.9, show_colorbar=True,
# title=str(lb[z[0]].cpu()),
# fig_size=(8, 10), use_pyplot=True)
# plotMe[0].savefig(str(number)+'NotEQLabel.jpg')
# ################################################
# outImg = img[z[0]].squeeze().detach().cpu().numpy()
# fig2 = plt.figure(figsize=(12,12))
# prImg = plt.imshow(outImg)
# fig2.savefig(str(number)+'NotEQOrig.jpg')
# ################################################
# fig = plt.figure(figsize=(15,10))
# ax = fig.add_subplot(111, projection='3d')
# z = upsampled_attr[z[0]].squeeze().detach().cpu().numpy()
# x = np.arange(0,64,1)
# y = np.arange(0,64,1)
# X, Y = np.meshgrid(x, y)
# plll = ax.plot_surface(X, Y , z, cmap=cm.coolwarm)
# # Customize the z axis.
# ax.set_zlim(np.min(z)+0.1*np.min(z),np.max(z)+0.1*np.max(z))
# ax.zaxis.set_major_locator(LinearLocator(10))
# ax.zaxis.set_major_formatter(FormatStrFormatter('%.02f'))
# # Add a color bar which maps values to colors.
# fig.colorbar(plll, shrink=0.5, aspect=5)
# fig.savefig(str(number)+'NotEQ3D.jpg')
#######explainVis####################
# if number%30 == 0:
# self.vis_explanation(number)
# self.vis_explanation(number)
#####################################
# reluMe = nn.ReLU(upsampled_attr)
# upsampled_attr = reluMe(upsampled_attr)
size_batch = upsampled_attr.size()[0]
# upsampled_attr = F.relu(upsampled_attr, inplace=False)
# print(upsampled_attr.size(),self.sintetic.size())
losssemantic = criteria2(
upsampled_attr[:size_batch, :, :, :],
self.sintetic[:size_batch, :, :, :].to(device))
loss1 = criteria1(out, lb)
# lossall = 0.7*loss1 + 0.3*losssemantic
lossall = losssemantic.to(device)
optim.zero_grad()
# loss2 = criteria21(upsampled_attr.squeeze()*64,maskLb)
# loss3 = criteriaL2(maskLb*upsampled_attr.squeeze(),maskLb*upsampled_attrB.squeeze())
# loss3 = criteriaL2(img.squeeze()*upsampled_attr.squeeze()*64,img.squeeze()*upsampled_attrB.squeeze()*64)
# loss3 = criteriaL2(img.squeeze()*upsampled_attr.squeeze(),img.squeeze()*upsampled_attrB.squeeze())
if number % 30 == 0:
# print()
print(loss1, losssemantic)
# loss3 = torch.log(-loss3)
# lossall = 0.7*loss1 + 0.3*loss2
# lossall = 0*loss1 + 0.3*loss3
# lossall = loss3
# print('Losss to cjeck is:--- ',torch.max(loss3))
avg_loss = torch.mean(lossall)
lossall.backward()
optim.step()
# print(avg_loss)
number += 1
# if number == 2:
# gradds = []
# for tag, parm in self.model.state_dict().items():
# if parm.requires_grad:
# # print(p.name, p.data)
# gradds.append(parm.grad.data.cpu().numpy())
# self.writer.add_histogram('grads', torch.from_numpy(gradds), number)
# self.writer.his
# self.plot_grad_flow(self.model.state_dict().items(),number)
if number % 10 == 0:
# print(number)
print(
"Epoch: {}/{} batch: {}/{} iteration: {}/{} average-loss: {:0.4f}"
.format(epoch + 1, n_epochs, i + 1,
len(self.KnownLoader), ite + 1, max_ite,
avg_loss.cpu()))
# self.writer.close()
# Save checkpoint
if (os.path.exists("./checkpoint")):
torch.save(self.model.state_dict(), "./checkpoint/resnet18.pth")
else:
os.mkdir('checkpoint')
torch.save(self.model.state_dict(), "./checkpoint/resnet18.pth")
number = 0
def attribute(self, img, target, **kwargs):
return LayerGradCam.interpolate(
CaptumDerivative.attribute(self, img, torch.tensor(target).cuda()),
img.shape[-2:])