နားလည်ထားရမယ့် အချက်
ဒီ course တစ်လျှောက် CNN classifier, transfer learning, object detection, segmentation (U-Net), self-supervised representation learning, edge deployment အတွက် model compression/quantization စတဲ့ tool အမျိုးမျိုးကို ကြည့်ခဲ့ပြီးဖြစ်တယ်။ Tool တစ်ခုစီက trade-off အသီးသီးရှိတယ် — data volume (labeled data များများ vs နည်းနည်း vs လုံးဝမရှိ)၊ output type (တစ်ပုံလုံးအတွက် label တစ်ခု vs object location တွေ vs pixel-level mask)၊ deployment constraint (server GPU vs low-power mobile chip) တို့ပေါ်မူတည်ပြီး ရွေးချယ်ရမယ်။ Architecture ရွေးတဲ့အခါ 'ဘယ်ဟာက accuracy အမြင့်ဆုံးလဲ' ဆိုတာထက် 'ဒီ problem ရဲ့ constraint တွေနဲ့ ဘယ်ဟာက အံဝင်ခွင်ကျဆုံးလဲ' ဆိုတာ စဉ်းစားရမယ်။
လက်တွေ့ scenario နဲ့ ချိတ်ကြည့်မယ်
Tutorial Platform ရဲ့ 'Architecture Picker' quiz widget ထဲမှာ ဒီလို scenario တွေကို random shuffle လုပ်ပြီး ပြပေးတယ် — user က architecture တစ်ခုစီအတွက် ဖြေဆိုပြီးရင် platform က reasoning rubric တစ်ခုနဲ့ compare လုပ်ပြီး ဘယ်အချက်တွေ လွဲနေလဲဆိုတာ feedback ပြန်ပေးတယ်။
အတူတူ စမ်းရေးကြည့်မယ်
# Read each scenario and decide which architecture/technique from
# this course fits best. This script just prints the scenarios --
# the actual exercise happens in your head (or on paper).
scenarios = [
{
"id": 1,
"description": "Classify product photos into 20 categories. "
"You have 50,000 labeled images.",
},
{
"id": 2,
"description": "Detect and count cars in real-time traffic "
"camera footage.",
},
{
"id": 3,
"description": "Segment tumors in CT scans, but only 200 "
"labeled scans are available.",
},
{
"id": 4,
"description": "Find visually similar images in a large photo "
"library with no labels at all.",
},
{
"id": 5,
"description": "Deploy an image classifier on a low-power "
"mobile device with strict latency limits.",
},
]
if __name__ == "__main__":
print("Computer Vision Architecture Picker")
print("=" * 40)
for scenario in scenarios:
print(f"Scenario {scenario['id']}: {scenario['description']}")
Script က scenarios list ကို header တစ်ကြောင်းနဲ့အတူ print ထုတ်ရုံပဲ လုပ်တယ် — model training, architecture ရွေးချယ်မှုစတာ ဘာမှ code ထဲမှာ မဖြစ်ပါဘူး။ Actual exercise က run ဖြစ်ပြီးနောက် reader ရဲ့ head ထဲမှာ ဖြစ်ပျက်ရမယ့် reasoning ဖြစ်တယ်။၅ မိနစ် စမ်းကြည့်
Script ကို run ပြီး scenario ငါးခုကို ဖတ်ပါ။ Scenario တစ်ခုစီအတွက် ဒီ course ထဲက architecture/technique (CNN classifier, transfer learning, object detection, segmentation/U-Net, self-supervised learning, edge deployment အတွက် compression/quantization) ထဲက ဘယ်ဟာက အသင့်တော်ဆုံးလဲဆိုတာ ရေးချပါ၊ ဘာကြောင့်လဲဆိုတာလည်း data volume, output type, deployment constraint တွေကို ကိုးကားပြီး ရှင်းပြပါ။
သတိလေးတစ်ချက်
Scenario တစ်ခုစီအတွက် 'အမြင့်ဆုံး accuracy ပေးမယ့် architecture' ကိုပဲ ရွေးချယ်ပြီး deployment constraint (mobile latency) ကို လျစ်လျူရှုတတ်တယ်။
Labeled data ပမာဏ (scenario 3 ရဲ့ 200 scans) ကို သတိမထားဘဲ data-hungry architecture ကို ရွေးမိတတ်တယ်၊ transfer learning/self-supervised approach ကို လျစ်လျူရှုတတ်တယ်။
Wikipedia — Computer vision — Computer Vision