QR Code Detection Model (#340)

analogdevicesinc · Jun 6, 2024 · 3b01ca3 · 3b01ca3
1 parent 53deb87
commit 3b01ca3
Show file tree

Hide file tree

Showing 7 changed files with 4,752 additions and 0 deletions.
diff --git a/networks/ai85-tinierssd-kpts-qr.yaml b/networks/ai85-tinierssd-kpts-qr.yaml
@@ -0,0 +1,304 @@
+---
+# HWC (little data) configuration for SVHN
+# Parallel Model
+
+arch: ai85tinierssdqr
+dataset: qrcode_160_120_kpts_ext
+
+layers:
+
+  # Layer 0: fire1, inp: 3x320x240
+  - out_offset: 0x1000
+    in_offset: 0x0000
+    processors: 0x0000000000000007
+    output_processors: 0xffffffff00000000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    streaming: true
+
+  # Layer 1: fire2, inp: 32x320x240
+  - out_offset: 0x1000
+    in_offset: 0x1000
+    processors: 0xffffffff00000000
+    output_processors: 0x00000000ffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    streaming: true
+
+  # Layer 2: fire3, inp: 32x320x240
+  - out_offset: 0x2440  # 0x2000
+    in_offset: 0x1000
+    processors: 0x00000000ffffffff
+    output_processors: 0xffffffffffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    max_pool: 2
+    pool_stride: 2
+    streaming: true
+
+  # Layer 3: fire3_1, inp: 64x160x120
+  - out_offset: 0x3000
+    in_offset: 0x2440
+    processors: 0xffffffffffffffff
+    output_processors: 0xffffffffffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    max_pool: 2
+    pool_stride: 2
+    streaming: true
+
+  # Layer 4: fire4, inp: 64x80x60
+  - out_offset: 0x4000
+    in_offset: 0x3000
+    processors: 0xffffffffffffffff
+    output_processors: 0xffffffffffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    streaming: true
+
+  # Layer 5: fire5, inp: 64x80x60
+  - out_offset: 0x5000  # 0x12C0
+    in_offset: 0x4000
+    processors: 0xffffffffffffffff
+    output_processors: 0xffffffffffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    max_pool: 2
+    pool_stride: 2
+    streaming: true
+
+  # Layer 6: fire6, inp: 64x40x30
+  - out_offset: 0x6400  # 0x12C0
+    in_offset: 0x5000
+    processors: 0xffffffffffffffff
+    output_processors: 0xffffffffffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+
+  # Layer 7: fire7, inp: 64x40x30
+  - out_offset: 0x0000  # 0x2580
+    in_offset: 0x6400
+    processors: 0xffffffffffffffff
+    output_processors: 0xffffffffffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+
+  # Layer 8: fire8, inp: 128x40x30
+  - name: fire8
+    out_offset: 0x3000  # 0x12C0
+    in_offset: 0x0000
+    processors: 0xffffffffffffffff
+    output_processors: 0x00000000ffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+
+  # Layer 9: fire9, inp: 32x40x30
+  - name: fire9
+    out_offset: 0x3000  # 0x04B0
+    in_offset: 0x3000
+    processors: 0x00000000ffffffff
+    output_processors: 0xffffffff00000000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    max_pool: 2
+    pool_stride: 2
+
+  # Layer 10: fire10, inp: 32x20x15
+  - name: fire10
+    out_offset: 0x4400  # 0x0280
+    in_offset: 0x3000
+    processors: 0xffffffff00000000
+    output_processors: 0x00000000ffffffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    max_pool: 2
+    pool_stride: 2
+
+  # Layer 11: conv12_1, inp: 32x10x7
+  - out_offset: 0x5000  # 0x0280
+    in_offset: 0x4400
+    processors: 0x00000000ffffffff
+    output_processors: 0x000000000000ffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+
+  # Layer 12: conv12_2, inp: 16x10x7
+  - name: conv12_2
+    out_offset: 0x5000  # 0x003C
+    in_offset: 0x5000
+    processors: 0x000000000000ffff
+    output_processors: 0x00000000ffff0000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    activate: ReLU
+    max_pool: 2
+    pool_stride: 2
+
+  # Layer 13: loc_conv8, inp: 32x40x30
+  - out_offset: 0x5400  # 0x12C0
+    in_offset: 0x3000
+    processors: 0x00000000ffffffff
+    output_processors: 0x000000000000ffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire8
+    output: true
+    activate: None
+
+  # Layer 14: loc_conv9, inp: 32x20x15
+  - out_offset: 0x66C0  # 0x04B0
+    in_offset: 0x3000
+    processors: 0xffffffff00000000
+    output_processors: 0x000000000000ffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire9
+    output: true
+    activate: None
+
+  # Layer 15: loc_conv10, inp: 32x10x7
+  - out_offset: 0x6B70  # 0x0118
+    in_offset: 0x4400
+    processors: 0x00000000ffffffff
+    output_processors: 0x000000000000ffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire10
+    output: true
+    activate: None
+
+  # Layer 16: loc_conv12_2, inp: 16x5x3
+  - out_offset: 0x6C88  # 0x003C
+    in_offset: 0x5000
+    processors: 0x00000000ffff0000
+    output_processors: 0x000000000000ffff
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: conv12_2
+    output: true
+    activate: None
+
+  # Layer 17: kpts_fire8, inp: 32x40x30
+  - out_offset: 0x4000  # 0x12C0
+    in_offset: 0x3000
+    processors: 0x00000000ffffffff
+    output_processors: 0xffffffff00000000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire8
+    output: true
+    activate: None
+
+  # Layer 18: kpts_fire9, inp: 32x20x15
+  - out_offset: 0x52C0  # 0x04B0
+    in_offset: 0x3000
+    processors: 0xffffffff00000000
+    output_processors: 0xffffffff00000000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire9
+    output: true
+    activate: None
+
+  # Layer 19: kpts_fire10, inp: 32x10x7
+  - out_offset: 0x5770  # 0x0118
+    in_offset: 0x4400
+    processors: 0x00000000ffffffff
+    output_processors: 0xffffffff00000000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire10
+    output: true
+    activate: None
+
+  # Layer 20: kpts_fire12_2, inp: 16x5x3
+  - out_offset: 0x5888
+    in_offset: 0x5000
+    processors: 0x00000000ffff0000
+    output_processors: 0xffffffff00000000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: conv12_2
+    output: true
+    activate: None
+
+  # Layer 21: cl_fire8, inp: 32x40x30
+  - out_offset: 0x5400  # 0x12C0
+    in_offset: 0x3000
+    processors: 0x00000000ffffffff
+    output_processors: 0x0000000000ff0000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire8
+    output: true
+    activate: None
+
+  # Layer 22: cl_conv9, inp: 32x20x15
+  - out_offset: 0x66C0  # 0x04B0
+    in_offset: 0x3000
+    processors: 0xffffffff00000000
+    output_processors: 0x0000000000ff0000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire9
+    output: true
+    activate: None
+
+  # Layer 23: cl_conv10, inp: 32x10x7
+  - out_offset: 0x6B70  # 0x0118
+    in_offset: 0x4400
+    processors: 0x00000000ffffffff
+    output_processors: 0x0000000000ff0000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: fire10
+    output: true
+    activate: None
+
+  # Layer 24: cl_conv12_2, inp: 16x5x3
+  - out_offset: 0x6C88  # 0x003C
+    in_offset: 0x5000
+    processors: 0x00000000ffff0000
+    output_processors: 0x0000000000ff0000
+    operation: conv2d
+    kernel_size: 3x3
+    pad: 1
+    in_sequences: conv12_2
+    output: true
+    activate: None
diff --git a/scripts/gen_qrcode_tinierssd.sh b/scripts/gen_qrcode_tinierssd.sh
@@ -0,0 +1,6 @@
+#!/bin/sh
+DEVICE="MAX78000"
+TARGET="sdk/Examples/$DEVICE/CNN"
+COMMON_ARGS="--device $DEVICE --timer 0 --display-checkpoint --verbose"
+
+python ai8xize.py --test-dir $TARGET --prefix tinierssd_kpts_qrcode --checkpoint-file trained/ai85-qrcode-tinierssd-kpts-qat8-q.pth.tar --config-file networks/ai85-tinierssd-kpts-qr.yaml --fifo $COMMON_ARGS --synthesize-input 0x2580  "$@"
diff --git a/scripts/quantize_qrcode_tinierssd.sh b/scripts/quantize_qrcode_tinierssd.sh
@@ -0,0 +1,2 @@
+#!/bin/sh
+python quantize.py trained/ai85-qrcode-tinierssd-qat8.pth.tar trained/ai85-qrcode-tinierssd-qat8-q.pth.tar --device MAX78000 -v
diff --git a/tests/sample_qrcode_160_120_kpts_ext.npy b/tests/sample_qrcode_160_120_kpts_ext.npy
diff --git a/trained/ai85-qrcode-tinierssd-kpts-qat8-q.pth.tar b/trained/ai85-qrcode-tinierssd-kpts-qat8-q.pth.tar