Add files using upload-large-folder tool

ground_truth/figma/analysis/prediction_results_gt_figma_analysis.json

@@ -0,0 +1,60 @@
+{
+  "total_samples": 28401,
+  "buckets": {
+    "[0.00MP, 2.00MP)": {
+      "count": 7993,
+      "accuracy": 0.5399724759164268,
+      "avg_latency": 0.7268637953522787,
+      "avg_tokens_per_s": 35.5162452017354,
+      "avg_gpu_mem": 7402.328070325695,
+      "avg_cpu_mem": -0.022190803671962966,
+      "avg_dist_error": 238.3239783386839,
+      "avg_text_tokens": 1170.3471787814337
+    },
+    "[2.00MP, 4.00MP)": {
+      "count": 5098,
+      "accuracy": 0.4727344056492742,
+      "avg_latency": 0.8131132628098232,
+      "avg_tokens_per_s": 32.28264258380644,
+      "avg_gpu_mem": 7715.044248357199,
+      "avg_cpu_mem": 0.0,
+      "avg_dist_error": 403.92873936088193,
+      "avg_text_tokens": 2720.205374656728
+    },
+    "[4.00MP, 6.00MP)": {
+      "count": 5163,
+      "accuracy": 0.4557427852024017,
+      "avg_latency": 0.971470576490209,
+      "avg_tokens_per_s": 27.347942534901378,
+      "avg_gpu_mem": 8218.006050885659,
+      "avg_cpu_mem": 0.0,
+      "avg_dist_error": 556.2626813390642,
+      "avg_text_tokens": 5222.537090838659
+    },
+    "[6.00MP, 8.00MP)": {
+      "count": 4934,
+      "accuracy": 0.46696392379408186,
+      "avg_latency": 1.146001877270361,
+      "avg_tokens_per_s": 23.33760377951412,
+      "avg_gpu_mem": 8732.531420809371,
+      "avg_cpu_mem": 0.0,
+      "avg_dist_error": 672.7381148162489,
+      "avg_text_tokens": 7781.48439400081
+    },
+    "[8.00MP, 10.00MP)": {
+      "count": 5213,
+      "accuracy": 0.43621714943410705,
+      "avg_latency": 1.360789297404775,
+      "avg_tokens_per_s": 19.806896554588153,
+      "avg_gpu_mem": 9285.341294824837,
+      "avg_cpu_mem": 0.12813968684059082,
+      "avg_dist_error": 782.8564262301205,
+      "avg_text_tokens": 10527.738730097832
+    }
+  },
+  "resize_impact": {},
+  "summary": {
+    "overall_accuracy": 0.48086334988204643,
+    "total_resized": 23507
+  }
+}

ground_truth/figma/analysis/prediction_results_gt_figma_analysis.txt

@@ -0,0 +1,238 @@
+
+==========================================================================================
+EXECUTIVE SUMMARY
+==========================================================================================
+Total samples analyzed: 28401
+Resized samples: 23507 (82.8%)
+Overall accuracy: 0.481
+Unresized accuracy: 0.483
+
+Best accuracy: 0.540 at 1.0MP
+Most efficient: 1.0MP (best accuracy per compute)
+
+PERFORMANCE BY PIXEL BUCKET
+==========================================================================================
+Pixel Bucket            Count   Accuracy   Latency(s)   Tokens/s    GPU(MB)
+------------------------------------------------------------------------------------------
+[0.00MP, 2.00MP)         7993      0.540        0.727       35.5     7402.3
+[2.00MP, 4.00MP)         5098      0.473        0.813       32.3     7715.0
+[4.00MP, 6.00MP)         5163      0.456        0.971       27.3     8218.0
+[6.00MP, 8.00MP)         4934      0.467        1.146       23.3     8732.5
+[8.00MP, 10.00MP)        5213      0.436        1.361       19.8     9285.3
+Unknown                     0      0.000        0.000        0.0        0.0
+
+RESIZE-ONLY PERFORMANCE BY PIXEL BUCKET
+==========================================================================================
+Pixel Bucket            Count   Accuracy   Latency(s)   Tokens/s    GPU(MB)
+------------------------------------------------------------------------------------------
+[0.00MP, 2.00MP)         4670      0.479        0.739       35.1     7451.2
+[2.00MP, 4.00MP)         4509      0.481        0.812       32.3     7706.6
+[4.00MP, 6.00MP)         4854      0.481        0.972       27.3     8218.3
+[6.00MP, 8.00MP)         4767      0.480        1.145       23.4     8732.1
+[8.00MP, 10.00MP)        4707      0.480        1.344       20.0     9253.3
+Unknown                     0      0.000        0.000        0.0        0.0
+
+NO-RESIZE PERFORMANCE BY PIXEL BUCKET
+==========================================================================================
+Pixel Bucket            Count   Accuracy   Latency(s)   Tokens/s    GPU(MB)
+------------------------------------------------------------------------------------------
+[0.00MP, 2.00MP)         3323      0.625        0.710       36.1     7333.6
+[2.00MP, 4.00MP)          589      0.409        0.822       31.8     7779.5
+[4.00MP, 6.00MP)          309      0.061        0.964       27.4     8213.9
+[6.00MP, 8.00MP)          167      0.084        1.169       22.8     8746.0
+[8.00MP, 10.00MP)         506      0.028        1.521       17.9     9583.3
+Unknown                     0      0.000        0.000        0.0        0.0
+
+RESIZE IMPACT ANALYSIS
+==========================================================================================
+
+UPSCALING (Lower → Higher Resolution)
+------------------------------------------------------------------------------------------
+Resize Direction        Count   Δ Accuracy   Δ Latency(s)   Δ Tokens/s
+------------------------------------------------------------------------------------------
+1MP→2MP                  3323       +0.009         -0.146         +3.4
+1MP→4MP                  3323       -0.001         +0.016         -1.6
+1MP→6MP                  3323       -0.009         +0.189         -5.6
+1MP→8MP                  3322       -0.019         +0.389         -9.0
+2MP→4MP                   589       +0.007         -0.020         -0.5
+2MP→6MP                   589       +0.007         +0.154         -4.5
+2MP→8MP                   589       +0.004         +0.347         -7.8
+4MP→6MP                   309       -0.001         +0.150         -4.3
+4MP→8MP                   309       +0.002         +0.348         -7.6
+6MP→8MP                   167       +0.005         +0.334         -7.2
+
+DOWNSCALING (Higher → Lower Resolution)
+------------------------------------------------------------------------------------------
+Resize Direction        Count   Δ Accuracy   Δ Latency(s)   Δ Tokens/s
+------------------------------------------------------------------------------------------
+2MP→1MP                   589       +0.021         -0.258         +7.5
+4MP→1MP                   309       +0.002         -0.262         +7.6
+4MP→2MP                   309       -0.001         -0.187         +4.8
+6MP→1MP                   167       -0.013         -0.270         +7.5
+6MP→2MP                   167       -0.013         -0.204         +4.9
+6MP→4MP                   167       +0.011         -0.043         -0.1
+8MP→1MP                   506       +0.002         -0.324         +8.3
+8MP→2MP                   506       -0.004         -0.255         +5.7
+8MP→4MP                   506       +0.008         -0.099         +0.9
+8MP→6MP                   506       -0.002         +0.073         -3.0
+
+SUMMARY:
+Upscaling operations: 10 resize patterns
+Downscaling operations: 10 resize patterns
+
+LOCALIZATION ERROR ANALYSIS
+==========================================================================================
+
+ABSOLUTE PIXEL ERROR - ALL PREDICTIONS
+Pixel Bucket            Count   Avg Error(px)    Median Error      Std Dev
+------------------------------------------------------------------------------------------
+[0.00MP, 2.00MP)         7993           238.3           139.6        291.3
+[2.00MP, 4.00MP)         5098           403.9           256.4        456.4
+[4.00MP, 6.00MP)         5163           556.3           382.5        598.8
+[6.00MP, 8.00MP)         4934           672.7           474.3        714.7
+[8.00MP, 10.00MP)        5213           782.9           578.2        783.5
+
+ABSOLUTE PIXEL ERROR - INCORRECT PREDICTIONS ONLY
+Pixel Bucket            Count   Avg Error(px)    Median Error      Std Dev
+------------------------------------------------------------------------------------------
+[0.00MP, 2.00MP)         3677           372.8           247.0        355.2
+[2.00MP, 4.00MP)         2688           588.0           410.3        525.7
+[4.00MP, 6.00MP)         2810           784.4           569.6        677.2
+[6.00MP, 8.00MP)         2630           948.3           678.6        810.8
+[8.00MP, 10.00MP)        2939          1064.7           840.3        861.3
+
+NORMALIZED ERROR RATIO (ERROR/IMAGE_DIMENSION) - ALL PREDICTIONS
+Pixel Bucket            Count     Avg Ratio %  Median Ratio %    Std Dev %
+------------------------------------------------------------------------------------------
+[0.00MP, 2.00MP)         7993           16.17           11.31        16.82
+[2.00MP, 4.00MP)         5098           16.71           11.27        17.28
+[4.00MP, 6.00MP)         5163           16.53           11.97        16.43
+[6.00MP, 8.00MP)         4934           16.27           11.92        16.13
+[8.00MP, 10.00MP)        5213           16.05           12.31        15.42
+
+NORMALIZED ERROR RATIO (%) - INCORRECT PREDICTIONS ONLY
+Pixel Bucket            Count     Avg Ratio %  Median Ratio %    Std Dev %
+------------------------------------------------------------------------------------------
+[0.00MP, 2.00MP)         3677           24.13           18.40        19.46
+[2.00MP, 4.00MP)         2688           23.76           17.63        19.24
+[4.00MP, 6.00MP)         2810           22.85           17.52        17.96
+[6.00MP, 8.00MP)         2630           22.45           17.27        17.82
+[8.00MP, 10.00MP)        2939           21.28           17.08        16.76
+
+SCALING LAWS ANALYSIS
+==========================================================================================
+Latency scaling: 0.691 × (megapixels)^0.29
+Memory scaling:  7271.1 × (megapixels)^0.11
+
+Interpretation:
+- Latency shows sublinear scaling (exponent=0.29)
+
+ILLUSTRATIVE EXAMPLES FOR MANUAL VERIFICATION
+==========================================================================================
+
+Bucket: [0.00MP, 2.00MP)
+------------------------------------------------------------
+  Example 1: 68154efe1c603b59f2a330da0f45907156294ec16c8812f250b1afac1f39868f.png:59
+    Result: ✗ INCORRECT
+    Original: 8.0MP → Resized: 1.0MP (Resized: YES)
+    Error: 2456.9 pixels
+    Latency: 0.798s
+
+  Example 2: 1c12e247f855483da55a78e3c3a8b59c3e445fbc754742c3f39486cbbe96ca68.png:18
+    Result: ✓ CORRECT
+    Original: 1.0MP → Resized: 1.0MP (Resized: YES)
+    Error: 365.5 pixels
+    Latency: 0.763s
+
+  Example 3: 79a4574c0e2daf55517f740fca6d7a25f58a0e08501f8c3105975e38c19fb65b.png:8
+    Result: ✓ CORRECT
+    Original: 1.0MP → Resized: 1.0MP (Resized: NO)
+    Error: 0.0 pixels
+    Latency: 0.701s
+
+
+Bucket: [2.00MP, 4.00MP)
+------------------------------------------------------------
+  Example 1: ffccdd3451081a3ea79feefd5dc91c12cfcd811894012a55114e733f6b2556fa.png:12
+    Result: ✗ INCORRECT
+    Original: 1.0MP → Resized: 2.0MP (Resized: YES)
+    Error: 3772.5 pixels
+    Latency: 0.866s
+
+  Example 2: cd4ff615e3c761ba8fe5d56cba99ab29655be8ee2ca07584a4169b97eb362c5b.png:20
+    Result: ✗ INCORRECT
+    Original: 1.0MP → Resized: 2.0MP (Resized: YES)
+    Error: 54.6 pixels
+    Latency: 0.853s
+
+  Example 3: cb88a6d44f4171bed27c02d2c87272448374eea2fb3be73d6e44a43d0b725687.png:10
+    Result: ✓ CORRECT
+    Original: 1.0MP → Resized: 2.0MP (Resized: YES)
+    Error: 1.0 pixels
+    Latency: 0.784s
+
+
+Bucket: [4.00MP, 6.00MP)
+------------------------------------------------------------
+  Example 1: ffccdd3451081a3ea79feefd5dc91c12cfcd811894012a55114e733f6b2556fa.png:12
+    Result: ✗ INCORRECT
+    Original: 1.0MP → Resized: 4.0MP (Resized: YES)
+    Error: 6395.7 pixels
+    Latency: 1.004s
+
+  Example 2: 73c08230a8f84d37ff1a0797cd4b5bd64fd23a292f60a18e364f0abe10532576.png:25
+    Result: ✗ INCORRECT
+    Original: 1.0MP → Resized: 4.0MP (Resized: YES)
+    Error: 107.5 pixels
+    Latency: 0.974s
+
+  Example 3: 88cbaf502f642ecc70a0bdc6c93a1033db1d692314950361f2672b6687e519b1.png:12
+    Result: ✓ CORRECT
+    Original: 2.0MP → Resized: 4.0MP (Resized: YES)
+    Error: 0.7 pixels
+    Latency: 0.931s
+
+
+Bucket: [6.00MP, 8.00MP)
+------------------------------------------------------------
+  Example 1: ffccdd3451081a3ea79feefd5dc91c12cfcd811894012a55114e733f6b2556fa.png:12
+    Result: ✗ INCORRECT
+    Original: 1.0MP → Resized: 6.0MP (Resized: YES)
+    Error: 8051.6 pixels
+    Latency: 1.197s
+
+  Example 2: 2bf430866443b4b6385b21ae729cd5170ee488273398d61cf61e034697db796c.png:13
+    Result: ✗ INCORRECT
+    Original: 1.0MP → Resized: 6.0MP (Resized: YES)
+    Error: 105.1 pixels
+    Latency: 1.165s
+
+  Example 3: 1969632f0966ea1b16a37c3bd35312aa3c1d25743741eb56c42bc7e03094db43.png:11
+    Result: ✓ CORRECT
+    Original: 1.0MP → Resized: 6.0MP (Resized: YES)
+    Error: 0.5 pixels
+    Latency: 1.153s
+
+
+Bucket: [8.00MP, 10.00MP)
+------------------------------------------------------------
+  Example 1: ffccdd3451081a3ea79feefd5dc91c12cfcd811894012a55114e733f6b2556fa.png:12
+    Result: ✗ INCORRECT
+    Original: 1.0MP → Resized: 8.0MP (Resized: YES)
+    Error: 9363.2 pixels
+    Latency: 1.369s
+
+  Example 2: 59b04b5656f3522a079a6e6fefee34b8fc5ea5eb64cf4a558254a58cd98df12f.png:51
+    Result: ✗ INCORRECT
+    Original: 1.0MP → Resized: 8.0MP (Resized: YES)
+    Error: 186.3 pixels
+    Latency: 1.348s
+
+  Example 3: 3b26e961108624bc41b7fc88a38d149f6d8d331cca7e433603822bd43eff01fa.png:13
+    Result: ✓ CORRECT
+    Original: 1.0MP → Resized: 8.0MP (Resized: YES)
+    Error: 0.0 pixels
+    Latency: 1.359s
+
+Total examples selected: 15
+==========================================================================================

ground_truth/figma/benchmark_summary_gt_figma.json

@@ -0,0 +1,62 @@
+{
+  "total_samples": 197244,
+  "correct": 100888,
+  "errors": 96779,
+  "accuracy": 0.5114883088965951,
+  "latency_avg_s": 1.8581367538401234,
+  "latency_p50_s": 1.3099228362552822,
+  "latency_p95_s": 4.30365471355617,
+  "avg_tokens_per_s": 20.227782101871668,
+  "avg_text_input_tokens": 4943.632744215287,
+  "avg_generated_tokens": 26.36999858043844,
+  "avg_gpu_mem_MB": 8161.776482035502,
+  "avg_cpu_mem_MB": -0.00046987563047954387,
+  "oom_samples": 0,
+  "resolution_stats": {
+    "1.0mpx": {
+      "samples": 57566,
+      "correct": 31591,
+      "accuracy": 0.548778793037557,
+      "avg_latency": 1.507000385774569
+    },
+    "2.0mpx": {
+      "samples": 35719,
+      "correct": 18201,
+      "accuracy": 0.5095607379825863,
+      "avg_latency": 1.629420851044059
+    },
+    "4.0mpx": {
+      "samples": 34964,
+      "correct": 17532,
+      "accuracy": 0.5014300423292529,
+      "avg_latency": 1.8826177604371819
+    },
+    "6.0mpx": {
+      "samples": 34204,
+      "correct": 16945,
+      "accuracy": 0.4954098935796983,
+      "avg_latency": 2.112791717425653
+    },
+    "8.0mpx": {
+      "samples": 34791,
+      "correct": 16619,
+      "accuracy": 0.47768100945646863,
+      "avg_latency": 2.39899063004073
+    }
+  },
+  "notes": {
+    "mpx_buckets": [
+      1.0,
+      2.0,
+      4.0,
+      6.0,
+      8.0
+    ],
+    "model_path": "/share_1/users/eddie/models/Qwen2.5-VL-3B-UI-R1-E",
+    "test_name": "gt_figma",
+    "dataset_name": "figma_jsonl",
+    "jsonl_path": "/share_1/users/eddie/data_creation/ground_truth/figma/layout400k_llm_processed.jsonl",
+    "images_dir": "/share_1/users/eddie/data/jedi/unpacked_images/figma400k",
+    "max_batch_size": 1
+  }
+}