File size: 10,778 Bytes
dbbceb8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
"""
Deep conv-stack detection head. Pure Conv2d — no reshaping overhead.

Pointwise (1x1) convs replace Linear layers.
Depthwise (3x3) convs provide spatial context.
Runs at full GPU throughput with fp16.
"""

import argparse
import json
import math
import os
import sys
import time

import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.cuda.amp import autocast, GradScaler

SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.insert(0, SCRIPT_DIR)

CACHE_DIR = os.environ.get("ARENA_CACHE_DIR")
COCO_ROOT = os.environ.get("ARENA_COCO_ROOT")
VAL_CACHE = os.environ.get("ARENA_VAL_CACHE")
RESOLUTION = 640
NUM_CLASSES = 80


def cofiber_decompose(f, n_scales):
    cofibers = []; residual = f
    for _ in range(n_scales - 1):
        omega = F.avg_pool2d(residual, 2)
        sigma_omega = F.interpolate(omega, size=residual.shape[2:], mode="bilinear", align_corners=False)
        cofibers.append(residual - sigma_omega); residual = omega
    cofibers.append(residual); return cofibers


class ConvBlock(nn.Module):
    """Pointwise conv + GELU + Depthwise spatial conv."""
    def __init__(self, in_ch, out_ch):
        super().__init__()
        self.pw = nn.Conv2d(in_ch, out_ch, 1)
        self.act = nn.GELU()
        self.dw = nn.Conv2d(out_ch, out_ch, 3, padding=1, groups=out_ch)
        self.norm = nn.GroupNorm(1, out_ch)  # instance norm per channel

    def forward(self, x):
        x = self.act(self.pw(x))
        x = self.norm(self.dw(x))
        return x


class DeepConvHead(nn.Module):
    """Pure conv-stack detection head on cofiber features."""

    def __init__(self, feat_dim=768, hidden=256, n_blocks=10, n_scales=3,
                 with_p3=False, lateral=False):
        super().__init__()
        self.n_scales = n_scales
        self.with_p3 = with_p3
        self.lateral = lateral
        n_total = n_scales + (1 if with_p3 else 0)
        self.scale_norms = nn.ModuleList([nn.GroupNorm(1, feat_dim) for _ in range(n_scales)])

        # Stem: project from feat_dim to hidden
        self.stem = nn.Conv2d(feat_dim, hidden, 1)
        self.stem_act = nn.GELU()

        # Stride-8 upsample path (P3)
        if with_p3:
            self.p3_upsample = nn.ConvTranspose2d(hidden, hidden, 2, stride=2)
            self.p3_norm = nn.GroupNorm(1, hidden)

        # Deep conv stack with residual connections
        self.blocks = nn.ModuleList()
        for _ in range(n_blocks):
            self.blocks.append(ConvBlock(hidden, hidden))

        # Lateral top-down fusion
        if lateral:
            self.lateral_convs = nn.ModuleList()
            self.lateral_norms = nn.ModuleList()
            for _ in range(n_scales - 1):
                self.lateral_convs.append(nn.Conv2d(hidden, hidden, 1))
                self.lateral_norms.append(nn.GroupNorm(1, hidden))

        # Output heads
        self.cls_head = nn.Conv2d(hidden, NUM_CLASSES, 1)
        self.reg_head = nn.Conv2d(hidden, 4, 1)
        self.ctr_head = nn.Conv2d(hidden, 1, 1)
        self.scale_params = nn.Parameter(torch.ones(n_total))

    def forward(self, spatial):
        cofibers = cofiber_decompose(spatial, self.n_scales)
        cls_l, reg_l, ctr_l = [], [], []

        # Process stride-16 first (needed for P3 upsample)
        scale_offset = 0
        if self.with_p3:
            cof16 = cofibers[0]  # stride 16, 40x40
            x16 = self.stem_act(self.stem(self.scale_norms[0](cof16)))
            for block in self.blocks:
                x16 = x16 + block(x16)

            # Create stride-8 via transposed conv (80x80)
            p3 = self.p3_norm(self.p3_upsample(x16))
            for block in self.blocks:
                p3 = p3 + block(p3)

            cls = self.cls_head(p3)
            reg_raw = (self.reg_head(p3) * self.scale_params[0]).clamp(-10, 10)
            reg = reg_raw.exp()
            ctr = self.ctr_head(p3)
            cls_l.append(cls); reg_l.append(reg); ctr_l.append(ctr)
            scale_offset = 1

        # Process each cofiber scale through the shared conv stack
        scale_features = []
        for i, cof in enumerate(cofibers):
            x = self.scale_norms[i](cof)
            x = self.stem_act(self.stem(x))
            for block in self.blocks:
                x = x + block(x)
            scale_features.append(x)

        # Top-down lateral fusion: coarse → fine
        if self.lateral:
            for i in range(len(scale_features) - 2, -1, -1):
                coarse = scale_features[i + 1]
                coarse_up = F.interpolate(coarse, size=scale_features[i].shape[2:],
                                          mode="bilinear", align_corners=False)
                scale_features[i] = self.lateral_norms[i](
                    scale_features[i] + self.lateral_convs[i](coarse_up))

        # Predict from each scale
        for i, x in enumerate(scale_features):
            cls = self.cls_head(x)
            reg_raw = (self.reg_head(x) * self.scale_params[i + scale_offset]).clamp(-10, 10)
            reg = reg_raw.exp()
            ctr = self.ctr_head(x)
            cls_l.append(cls); reg_l.append(reg); ctr_l.append(ctr)
        return cls_l, reg_l, ctr_l


def make_locations(feature_sizes, strides, device):
    locs = []
    for (h, w), s in zip(feature_sizes, strides):
        ys = (torch.arange(h, device=device, dtype=torch.float32) + 0.5) * s
        xs = (torch.arange(w, device=device, dtype=torch.float32) + 0.5) * s
        gy, gx = torch.meshgrid(ys, xs, indexing="ij")
        locs.append(torch.stack([gx.flatten(), gy.flatten()], -1))
    return locs


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--hidden", type=int, default=256)
    parser.add_argument("--blocks", type=int, default=10)
    parser.add_argument("--epochs", type=int, default=8)
    parser.add_argument("--batch-size", type=int, default=64)
    parser.add_argument("--lr", type=float, default=1e-3)
    parser.add_argument("--with-p3", action="store_true", help="Add stride-8 level via transposed conv")
    parser.add_argument("--lateral", action="store_true", help="Top-down lateral connections between scales")
    args = parser.parse_args()

    head = DeepConvHead(hidden=args.hidden, n_blocks=args.blocks, with_p3=args.with_p3,
                        lateral=args.lateral).cuda()
    n_params = sum(p.numel() for p in head.parameters())
    print("=" * 60)
    print(f"Deep Conv Head: {args.hidden} hidden, {args.blocks} blocks")
    print(f"  {n_params:,} params")
    print("=" * 60, flush=True)

    from cache_and_train_fast import compute_loss
    manifest = json.load(open(os.path.join(CACHE_DIR, "manifest.json")))
    n_shards = manifest["n_shards"]
    n_images = manifest["n_images"]
    steps_per_epoch = n_images // args.batch_size
    total_steps = steps_per_epoch * args.epochs
    warmup = int(total_steps * 0.03)

    optimizer = torch.optim.AdamW(head.parameters(), lr=args.lr, weight_decay=1e-4)
    scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lambda s:
        s / max(warmup, 1) if s < warmup else
        0.5 * (1 + math.cos(math.pi * (s - warmup) / max(total_steps - warmup, 1))))
    scaler = GradScaler()

    H = RESOLUTION // 16
    if args.with_p3:
        strides = [8, 16, 32, 64]
        locs = make_locations([(H*2,H*2),(H,H),(H//2,H//2),(H//4,H//4)], strides, torch.device("cuda"))
    else:
        strides = [16, 32, 64]
        locs = make_locations([(H,H),(H//2,H//2),(H//4,H//4)], strides, torch.device("cuda"))
    shard_paths = [os.path.join(CACHE_DIR, f"shard_{i:04d}.pt") for i in range(n_shards)]

    print(f"  {n_images} images, batch {args.batch_size}, {total_steps} steps, {args.epochs} epochs")
    print(f"  fp16 mixed precision enabled")
    print(f"  Training...\n", flush=True)

    head.train()
    global_step = 0
    t0 = time.time()

    for epoch in range(args.epochs):
        shard_order = torch.randperm(n_shards).tolist()
        epoch_t0 = time.time()

        for shard_idx in shard_order:
            if global_step >= total_steps: break
            shard = torch.load(shard_paths[shard_idx], map_location="cpu", weights_only=False)
            within = torch.randperm(len(shard)).tolist()

            for batch_start in range(0, len(shard), args.batch_size):
                if global_step >= total_steps: break
                batch_idx = within[batch_start:batch_start + args.batch_size]
                if len(batch_idx) < 2: continue

                spatial = torch.stack([shard[i]["spatial"] for i in batch_idx]).float().cuda()
                boxes = [shard[i]["boxes"].cuda() for i in batch_idx]
                labels = [shard[i]["labels"].cuda() for i in batch_idx]

                try:
                    with autocast():
                        cls_l, reg_l, ctr_l = head(spatial)
                        loss = compute_loss(cls_l, reg_l, ctr_l, locs, boxes, labels)

                    optimizer.zero_grad()
                    scaler.scale(loss).backward()
                    scaler.unscale_(optimizer)
                    torch.nn.utils.clip_grad_norm_(head.parameters(), 5.0)
                    scaler.step(optimizer)
                    scaler.update()
                    scheduler.step()

                    global_step += 1

                    if global_step % 100 == 0:
                        lr = scheduler.get_last_lr()[0]
                        elapsed = time.time() - t0
                        print(f"  step {global_step}/{total_steps} (ep {epoch+1}) "
                              f"loss={loss.item():.4f} lr={lr:.2e} "
                              f"{global_step/elapsed:.1f} it/s", flush=True)

                    if global_step % 4000 == 0:
                        ckpt = f"/home/zootest/checkpoint_convdeep_step{global_step}.pth"
                        torch.save({"head": head.state_dict(), "step": global_step}, ckpt)

                except RuntimeError as e:
                    if "out of memory" in str(e):
                        torch.cuda.empty_cache()
                        optimizer.zero_grad()
                        global_step += 1
                        scheduler.step()
                        continue
                    raise

            del shard

        print(f"  Epoch {epoch+1}/{args.epochs} complete ({time.time()-epoch_t0:.0f}s)\n", flush=True)

    # Save
    out_dir = os.path.join(SCRIPT_DIR, "heads", "cofiber_threshold", "conv_deep")
    os.makedirs(out_dir, exist_ok=True)
    out = os.path.join(out_dir, f"conv_deep_{args.hidden}h_{args.blocks}b_{args.epochs}ep.pth")
    torch.save(head.state_dict(), out)
    elapsed = time.time() - t0
    print(f"Saved: {out}")
    print(f"{n_params:,} params, {elapsed/60:.1f} minutes")


if __name__ == "__main__":
    main()