Hugging Face's logo

Spaces:
dheepakkaran
/
multi-camera-bev
Sleeping

App Files Community
multi-camera-bev
File size: 4,833 Bytes
d78fec2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# ============================================================
# constants.py
# ============================================================
#
# ETHUKU CREATE PANNINOM? (Why does this file exist?)
# ━━━━━━━━━━━━━━━━━━━━━━
# Project full-la ellarum use panra "magic numbers" oru edathula
# irukkanum. Image size, camera names, BEV grid size, class names.
# Ithu illaina ovvoru file-layum 224, 400 nu type panni, oru naal
# oru edathula mattum maathina bug varum.
#
# MUNADI FILE ODA CONNECTION: (How does it connect to other files?)
# ━━━━━━━━━━━━━━━━━━━━━━━━━━
# First file. Ithu yaaraiyum import pannala.
# Ithuku aprom varra ellathum (camera_loader, dataset, models) ithai
# import pannum.
#
# INNER OPERATIONS: (What happens inside?)
# ━━━━━━━━━━━━━━━━
# Vera onnum illa - just constant values define pannurom.
#
# INPUT / OUTPUT:
# ━━━━━━━━━━━━━━
# Input: illa. Output: module-level constants.
#
# EPADI USE AAGUM: (How is it used in the bigger system?)
# ━━━━━━━━━━━━━━━
# from data.scripts.constants import TARGET_W, TARGET_H, CAMERAS
#
# ============================================================

# --- Image size ---
# nuScenes original photo = 1600 x 900 pixels (romba periyathu).
# Athai 400 x 224 ku suruki-kirom -> 16x less pixels -> fast training.
# 224 & 400 ellam 32-la divide aagum (CNN downsample 32x pannum,
# so remainder illama irukkanum).
ORIGINAL_W = 1600
ORIGINAL_H = 900
TARGET_W = 400
TARGET_H = 224

# Resize scale factor. Camera K matrix-yum ithe scale-la maathanum,
# illaina 3D math thappa poidum.
SCALE_W = TARGET_W / ORIGINAL_W   # 0.25
SCALE_H = TARGET_H / ORIGINAL_H   # 0.2489

# --- 6 cameras (order MUKIYAM, always same order) ---
# Car mela 6 camera. Order fix panniten - model ithe order-la
# ethirpaakkum.
CAMERAS = [
    "CAM_FRONT",
    "CAM_FRONT_RIGHT",
    "CAM_BACK_RIGHT",
    "CAM_BACK",
    "CAM_BACK_LEFT",
    "CAM_FRONT_LEFT",
]
N_CAMERAS = len(CAMERAS)   # 6

# --- ImageNet normalization ---
# EfficientNet-B0 ImageNet photos-la pretrain aagirukku.
# Anga use panna mean/std ithu. Same normalization pannina thaan
# pretrained weights correct-a velai seiyum.
IMAGENET_MEAN = [0.485, 0.456, 0.406]
IMAGENET_STD = [0.229, 0.224, 0.225]

# --- BEV grid (top-down map) ---
# Car center-la nikkuthu. Suthi 100m x 100m area-va
# 200 x 200 cells-a pirikirom. Oru cell = 0.5m x 0.5m.
BEV_H = 200
BEV_W = 200
BEV_RESOLUTION = 0.5          # metres per cell
X_RANGE = (-50.0, 50.0)       # ego x = pinnadi(-)/munnadi(+), metres
Y_RANGE = (-50.0, 50.0)       # ego y = valathu(-)/idathu(+), metres
Z_RANGE = (-5.0, 3.0)         # ego z = keezha(-)/mela(+), metres
# nuScenes ego frame: x munnadi, y idathu pakkam, z mela. Ithu
# LSS-um dataset target-um ORE frame use pannurathunala thaan
# box position-um camera projection-um match aaguthu.

# --- LSS depth bins ---
# Camera-la depth theriyathu. So "2m to 50m varaikum 64 guesses"
# nu vachi, ovvoru guess-kum probability predict pannuvom.
D_MIN = 2.0
D_MAX = 50.0
N_DEPTHS = 64

# --- Channels ---
BACKBONE_OUT_CHANNELS = 64    # camera feature channels
BEV_OUT_CHANNELS = 128        # BEV encoder output channels

# --- 10 nuScenes detection classes ---
CLASSES = [
    "car",
    "truck",
    "bus",
    "trailer",
    "construction_vehicle",
    "pedestrian",
    "motorcycle",
    "bicycle",
    "traffic_cone",
    "barrier",
]
N_CLASSES = len(CLASSES)   # 10

# nuScenes-la category name romba long ("vehicle.car").
# Athai namma 10 class-ku map pannurom.
NUSCENES_NAME_MAP = {
    "vehicle.car": "car",
    "vehicle.truck": "truck",
    "vehicle.bus.bendy": "bus",
    "vehicle.bus.rigid": "bus",
    "vehicle.trailer": "trailer",
    "vehicle.construction": "construction_vehicle",
    "human.pedestrian.adult": "pedestrian",
    "human.pedestrian.child": "pedestrian",
    "human.pedestrian.construction_worker": "pedestrian",
    "human.pedestrian.police_officer": "pedestrian",
    "vehicle.motorcycle": "motorcycle",
    "vehicle.bicycle": "bicycle",
    "movable_object.trafficcone": "traffic_cone",
    "movable_object.barrier": "barrier",
}
CLASS_TO_IDX = {name: i for i, name in enumerate(CLASSES)}

# --- Dataset paths / split ---
DATA_ROOT = "data/nuscenes-mini"
VERSION = "v1.0-mini"

# nuScenes OFFICIAL mini split use pannurom: 8 train scene, 2 val scene.
# Yaen official? Official NDS evaluation "mini_val" scene list-ai
# ethirpaakkum. Namma sonthama split panna, official metric run panna
# mudiyaathu (apo resume-la NDS number podave mudiyaathu).
# Same scene train+val la irukka koodathu (illaina model mugam
# paathurum = cheating) - official split athai already kavanichirukku.
VAL_SCENES = ["scene-0103", "scene-0916"]