-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathvis.py
More file actions
172 lines (153 loc) · 11.9 KB
/
vis.py
File metadata and controls
172 lines (153 loc) · 11.9 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
import os,io,shutil
import geometry
import wandb
from matplotlib import cm
import cv2
from tqdm import tqdm
import torchvision
import time
from torchvision.utils import make_grid,draw_keypoints
import torch.nn.functional as F
import kornia
import numpy as np
import torch
import flow_vis
import flow_vis_torch
import matplotlib.pyplot as plt
from einops import rearrange, repeat
import models
import piqa
import imageio
from PIL import Image
#import splines.quaternion
#from torchcubicspline import (natural_cubic_spline_coeffs, NaturalCubicSpline)
from scipy import spatial
import plotly.express as px
import plotly.graph_objects as go
from collections import defaultdict
ch_fst = lambda src,x=None:rearrange(src,"... (x y) c -> ... c x y",x=int(src.size(-2)**(.5)) if x is None else x)
ch_sec = lambda x: rearrange(x,"... c x y -> ... (x y) c")
def wandb_summary(loss, model_output, model_input, ground_truth, resolution,prefix="",suffix="",step=0,losses_agg=[]):
model_output,model_input,ground_truth = [{k:(v[:32] if len(v.shape) else v) for k,v in x.items()} for x in (model_output,model_input,ground_truth)]
resolution = list(model_input["rgb"].flatten(0,1).permute(0,2,3,1).shape)
resolution[0]=ground_truth["rgb"].size(1)*ground_truth["rgb"].size(0)
nrow=model_input["rgb"].size(1)
imsl=model_input["rgb"].shape[-2:]
inv = lambda x : 1/(x+1e-8)
# Convert depths to colormapped 3-channel images:
for k,v in list(model_output.items()): # magma colormap for depth -- todo change to depth_colored instead of depth to avoid ambiguity
if type(v)!=list and len(v.shape): v=v.clip(min=.001)
if "depth" in k: model_output[k+"_raw"] = v
#if "depth" in k and "raw" not in k: model_output[k+"vis"] = v.expand(-1,-1,-1,3)#torch.from_numpy(cm.get_cmap('magma')(v.min().item()/v.cpu().numpy())).squeeze(-2)[...,:3]
if "depth" in k and "raw" not in k: model_output[k+"vis"] = torch.from_numpy(cm.get_cmap('magma')(v.min().item()/v.cpu().numpy())).squeeze(-2)[...,:3]
wandb_out = {}
#if step%50==0:
wandb_out["ref/rgb_gt"]= make_grid(model_input["rgb"].cpu().flatten(0,1).detach()*.5+.5,nrow=nrow)
if "lie_perpix" in model_output:
rot_vis=kornia.geometry.conversions.quaternion_to_axis_angle(model_output["lie_perpix"][...,:4]).flatten(0,1).permute(0,3,1,2)*.5+.5
trans_vis = model_output["lie_perpix"][...,-3:].flatten(0,1).permute(0,3,1,2)/5+.5
wandb_out["est/poses_lie_rot_perpix"]= make_grid(rot_vis.detach(),nrow=nrow,normalize=False)
wandb_out["est/poses_lie_trans_perpix"]= make_grid(trans_vis.detach(),nrow=nrow,normalize=False)
if "rig_masks" in model_output:
low_res=imsl#(64,64)
wandb_out["est/rig_masks"]= make_grid(rearrange(model_output["rig_masks"],"b t o (x y) 1 -> (b t o) 1 x y",x=low_res[0]).detach(),nrow=model_output["rig_masks"].size(2))
wandb_out["est/rig_masks_corr_weighted"]= make_grid((F.interpolate(model_output["corr_weights"][0],low_res).unsqueeze(2)*ch_fst(model_output["rig_masks"][0,1:],low_res[0])
).flatten(0,1).detach(),nrow=model_output["rig_masks"].size(2))
wandb_out["est/rig_masks_corr_weighted_rgb"]= make_grid((F.interpolate(model_input["rgb"][0,1:],low_res).unsqueeze(1)*ch_fst(model_output["rig_masks"][0,1:],low_res[0])*
F.interpolate(model_output["corr_weights"][0],low_res).unsqueeze(2)
).flatten(0,1).detach(),nrow=model_output["rig_masks"].size(2))
wandb_out["est/rig_masks_rgb"]= make_grid(rearrange(model_output["rig_masks"].flatten(0,1)*(F.interpolate(model_input["rgb"].flatten(0,1),low_res).flatten(-2,-1).permute(0,2,1).unsqueeze(1)*.5+.5),
"bt o (x y) c -> (bt o) c x y",x=low_res[0]).detach(),nrow=model_output["rig_masks"].size(2))
if "depth_inpvis" in model_output: wandb_out["est/depth_inp"]=make_grid(model_output["depth_inpvis"].cpu().flatten(0,1).permute(0,2,1).unflatten(-1,imsl).detach(),nrow=nrow)
if "res_depthvis" in model_output: wandb_out["est/res_depth"]=make_grid(model_output["res_depthvis"].cpu().flatten(0,1).permute(0,2,1).unflatten(-1,imsl).detach(),nrow=nrow)
if "depthvis" in model_output: wandb_out["est/depth"]=make_grid(model_output["depthvis"].cpu().flatten(0,1).permute(0,2,1).unflatten(-1,imsl).detach(),nrow=nrow)
if "depth_raw" in model_output:
wandb_out["est/depth_raw"]=make_grid(model_output["depth_raw"].cpu().flatten(0,1).permute(0,2,1).unflatten(-1,imsl).detach(),nrow=nrow,normalize=True)
wandb_out["est/depth_raw_inv"]=make_grid(inv(model_output["depth_raw"]).cpu().flatten(0,1).permute(0,2,1).unflatten(-1,imsl).detach(),nrow=nrow,normalize=True)
if "corr_weights" in model_output: wandb_out["est/corr_weights"] = make_grid(model_output["corr_weights"].flatten(0,1).cpu().detach(),normalize=True,nrow=nrow)
if "corr_weights_static" in model_output: wandb_out["est/corr_weights_static"] = make_grid(model_output["corr_weights_static"].flatten(0,1).cpu().detach(),normalize=True,nrow=nrow)
if "bwd_flow" in model_input: wandb_out["ref/flow_gt_bwd"]= flow_vis_torch.flow_to_color(make_grid(model_input["bwd_flow"].flatten(0,1),nrow=nrow-1))/255
if "rig_flow_masks" in model_input:
wandb_out["ref/rig_flow_masks"]= make_grid(rearrange(model_input["rig_flow_masks"],"b t o x y -> (b t o) 1 x y"),nrow=model_input["rig_flow_masks"].size(1))
if "flow_from_pose" in model_output and not torch.isnan(model_output["flow_from_pose"]).any():
wandb_out["est/flow_est_pose"] = flow_vis_torch.flow_to_color(make_grid(model_output["flow_from_pose"].clip(-.1,.1).flatten(0,1).permute(0,2,1).unflatten(-1,imsl),nrow=nrow-1))/255
if "affinity_emb" in model_output:
for suff in ["","_unnorm"][:1]:
aff_emb = model_output["affinity_emb"+suff]
if aff_emb.size(2)<3: aff_emb = torch.cat((aff_emb,torch.zeros_like(aff_emb[:,:,[0]]).expand(-1,-1,3-aff_emb.size(2),-1,-1)),2)
features=rearrange(aff_emb.flatten(0,1),"bt c x y -> 1 c (bt x) y")
B, C, H, W = features.shape
features = features.view(B, C, -1)
# Center the data
features_mean = features.mean(dim=2, keepdim=True)
features = features - features_mean
covariance = torch.bmm(features, features.transpose(1, 2)) / (H * W - 1)
# Perform SVD
U, S, V = torch.svd(covariance)
# Project the data onto the top principal components
num_components=min(6,C)
transformed_features = torch.bmm(U[:, :, :num_components].transpose(1, 2), features)
# Reshape back to original spatial dimensions
wandb_out["est/affinity_emb0to3"+suff]= make_grid(rearrange(transformed_features[:,:3].detach(),"1 c (bt x y) -> (bt) c x y",y=aff_emb.size(-1),x=aff_emb.size(-2)), nrow=model_output["affinity_emb"].size(1),normalize=True)
wandb_out["est/affinity_emb3to6"+suff]= make_grid(rearrange(transformed_features[:,3:].detach(),"1 c (bt x y) -> (bt) c x y",y=aff_emb.size(-1),x=aff_emb.size(-2)), nrow=model_output["affinity_emb"].size(1),normalize=True)
# Visualize point track reprojection error
if "pred_tracks" in model_input:
# Plot tracks as flow image
sl=64 if model_input["pred_tracks"].size(-2)%64**2==0 else 42
nrow_=int(model_input["pred_tracks"].size(-2)//sl**2)
low_res=(sl,sl)
uv = np.mgrid[0 : low_res[0], 0 : low_res[1]].astype(float).transpose(1, 2, 0)
uv = torch.from_numpy(np.flip(uv, axis=-1).copy()).long()
uv = uv / torch.tensor([low_res[1]-1, low_res[0]-1]) # uv in [0,1]
track_unp = lambda x: rearrange(x,"b t (x y s) c -> (b t s) c x y",y=sl,x=sl)
wandb_out["ref/track_flow_gt"] = flow_vis_torch.flow_to_color(make_grid((track_unp(model_input["pred_tracks"]*model_input["pred_visibility"].unsqueeze(-1)) -
uv.permute(2,0,1)[None].cuda())*track_unp(model_input["pred_visibility"].unsqueeze(-1)),nrow=nrow_))/255
if "aff_sim_grid" in model_output:
wandb_out["est/aff_grid"] = make_grid(model_output["aff_sim_grid"].flatten(0,2)[:,None],nrow=16)
if "point_track_reproj" in model_output:
# First just error image
point_track_err = ( (model_output["point_track_reproj"] - model_input["pred_tracks"]) * model_input["pred_visibility"].unsqueeze(-1) ).abs()
err_img = make_grid(rearrange(point_track_err,"b t (x y s) c -> (b t s) c x y",y=sl,x=sl),nrow=nrow_)*4#/255
wandb_out["metrics/track_err_x"] = err_img[[0]].expand(3,-1,-1)
wandb_out["metrics/track_err_y"] = err_img[[1]].expand(3,-1,-1)
wandb_out["est/track_flow_est"] = flow_vis_torch.flow_to_color(make_grid((track_unp(model_output["point_track_reproj"]*model_input["pred_visibility"].unsqueeze(-1)) -
uv.permute(2,0,1)[None].cuda())*track_unp(model_input["pred_visibility"].unsqueeze(-1)),nrow=nrow_))/255
if "rig_pertrack" in model_output: wandb_out["est/rig_samps"] = make_grid(track_unp(model_output["rig_pertrack"][None,...,None]),nrow=nrow_)
if "aff_emb_pertrack" in model_output:
features = rearrange(model_output["aff_emb_pertrack"],"b (x y s) c -> s c (b x y)",y=sl,x=sl)[[0]]
# Center the data
features_mean = features.mean(dim=2, keepdim=True)
features = features - features_mean
covariance = torch.bmm(features, features.transpose(1, 2)) / (features.size(-1) - 1)
# Perform SVD
U, S, V = torch.svd(covariance)
# Project the data onto the top principal components
num_components=min(3,features.size(1))
transformed_features = torch.bmm(U[:, :, :num_components].transpose(1, 2), features)
# Reshape back to original spatial dimensions
wandb_out["est/affinity_emb_track"]= make_grid(rearrange(transformed_features.detach(),"1 c (b x y) -> b c x y",y=sl,x=sl), normalize=True)
if "poses_all" in model_output:
poses_lie = torch.cat((kornia.geometry.conversions.rotation_matrix_to_quaternion(model_output["poses_all"][...,:3,:3],eps=1e-5),model_output["poses_all"][...,:3,-1]),-1)
rot_vis=kornia.geometry.conversions.quaternion_to_axis_angle(poses_lie[...,:4])*.5+.5
trans_vis = poses_lie[...,-3:]/5+.5
wandb_out["est/poses_lie_rot_all"]= make_grid( rearrange(rot_vis.detach(),"b (x y s) t c -> (b s t) c x y ",y=sl,x=sl), nrow=poses_lie.size(2),normalize=False)
wandb_out["est/poses_lie_trans_all"]= make_grid( rearrange(trans_vis.detach(),"b (x y s) t c -> (b s t) c x y ",y=sl,x=sl), nrow=poses_lie.size(2),normalize=False)
rgbcamimgs = torch.stack((
rearrange(model_output["rgb_pertrack"].detach(),"b (x y s) c -> b s c x y ",y=sl,x=sl)[:,0]*.5+.5,
rearrange(rot_vis.detach(),"b (x y s) t c -> b s t c x y ",y=sl,x=sl)[:,0,-1],
rearrange(trans_vis.detach(),"b (x y s) t c -> b s t c x y ",y=sl,x=sl)[:,0,-1],
),1).flatten(0,1)
wandb_out["est/rgb_camimgs"]= make_grid( rgbcamimgs, nrow=3,normalize=False)
#plt.imsave("/home/cameronsmith/tmp.png",wandb_out["est/rgb_camimgs"].permute(1,2,0).cpu().numpy())
if 0: # save locally
for k,v in wandb_out.items(): print(k,v.max(),v.min())
for k,v in wandb_out.items():
print(k,v.shape)
plt.imsave("output/img/%s.png"%k,v.float().permute(1,2,0).detach().cpu().numpy().clip(0,1));
print("saving locally")
zz
print("logging images",print(len(wandb_out)))
for k,v in wandb_out.items():print(k,v.shape)
wandb.log({prefix+k:wandb.Image(v.permute(1, 2, 0).float().detach().clip(0,1).cpu().numpy()) for k,v in wandb_out.items()})
print("done logging images")
return wandb_out