Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Targets rebuilt so the picture a person reviews is the picture the model is trained and scored on. Every V4.1 target is RGBA and PIL's convert("RGB") -- used by the benchmark, the training image loader, the vLLM inference path and core/vision.py -- drops alpha and keeps the RGB stored underneath, while a browser composites.
- 61 targets hid a neighbouring widget under the mask; those pixels are now white. On them the pool's own reference render scored SSIM 0.5532 against the stored target and 0.6990 against the flattened one, against a pool mean of 0.7664. White was measured against the alternatives: white 0.7163, unchanged 0.6161, black 0.5557.
- 108 targets are cropped to the bounding box of their non-transparent pixels; a transparent margin is what the capture left, not part of the design. Opaque white margin is kept.
- side_info regenerated from the new pixels with w2c-bench 1.2.0 on CPU; sha256 updated; eval dropped as a stale ground-truth cache; flattened_from records the source…
Every `rendered.png` is now the output of houstonzhang/w2c-render:1.3.1, and
238 of the 2,822 changed. The other 2,584 are byte-identical to what was here.
The old renderer did not draw the same source the same way twice. Chromium
reused a tile from an earlier frame when only part of it was invalidated, so
about one chart render in two hundred came back a different picture; and a
chart that did not disable its own animation -- recharts 3 leaves that on by
default -- was screenshotted at whichever frame the clock allowed. 236 of the
changes are the first cause and are sub-pixel antialiasing: the worst is SSIM
0.999451, the median 0.999999, and none reach the 0.999 gate the SFT build
applies. Two are a `<ReferenceLine>` returning to the paint order an
unperturbed first render gives it, worth 5 and 8 pixels.
One is not cosmetic. `test/image_1873` leaves `isAnimationActive` off one of
its two series, so the old image was a frame of a running animation: the same
source produced six differ…
