Datasets:

victorzhu30
/

flashvsr-repro-outputs-part2

source string	codec string	ffv1_args list	sequences list	failed list	src_bytes int64	mkv_bytes int64
outputs/reds_val30_gate_frame	ffv1	[ "-c:v", "ffv1", "-level", "3", "-coder", "1", "-context", "1", "-g", "1", "-slices", "4", "-slicecrc", "1", "-pix_fmt", "rgb24" ]	[ { "seq": "000", "frames": 93, "width": 1280, "height": 640, "names": [ "00000000.png", "00000001.png", "00000002.png", "00000003.png", "00000004.png", "00000005.png", "00000006.png", "00000007.png", "00000008.png", "00000009.png", ...	[]	3,705,442,837	2,528,161,833
outputs/reds_val30_gate_lam20_fifo	ffv1	[ "-c:v", "ffv1", "-level", "3", "-coder", "1", "-context", "1", "-g", "1", "-slices", "4", "-slicecrc", "1", "-pix_fmt", "rgb24" ]	[ { "seq": "000", "frames": 93, "width": 1280, "height": 640, "names": [ "00000000.png", "00000001.png", "00000002.png", "00000003.png", "00000004.png", "00000005.png", "00000006.png", "00000007.png", "00000008.png", "00000009.png", ...	[]	3,707,802,550	2,531,612,006
outputs/reds_val30_gate_lam5	ffv1	[ "-c:v", "ffv1", "-level", "3", "-coder", "1", "-context", "1", "-g", "1", "-slices", "4", "-slicecrc", "1", "-pix_fmt", "rgb24" ]	[ { "seq": "000", "frames": 93, "width": 1280, "height": 640, "names": [ "00000000.png", "00000001.png", "00000002.png", "00000003.png", "00000004.png", "00000005.png", "00000006.png", "00000007.png", "00000008.png", "00000009.png", ...	[]	3,706,731,367	2,528,830,875
outputs/reds_val30_reliability_v	ffv1	[ "-c:v", "ffv1", "-level", "3", "-coder", "1", "-context", "1", "-g", "1", "-slices", "4", "-slicecrc", "1", "-pix_fmt", "rgb24" ]	[ { "seq": "000", "frames": 93, "width": 1280, "height": 640, "names": [ "00000000.png", "00000001.png", "00000002.png", "00000003.png", "00000004.png", "00000005.png", "00000006.png", "00000007.png", "00000008.png", "00000009.png", ...	[]	3,708,285,620	2,531,654,378

YAML Metadata Warning:The task_categories "video-super-resolution" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other

FlashVSR 复现实验输出 — part2

FlashVSR 复现及 KV cache 驱逐策略消融实验的逐帧推理输出，以 FFV1 无损编码归档。本 repo 是全部结果的第 2/3 部分。

内容

reds_bscv_sliding_kv10
reds_inter4k30_ofr_headwise
reds_inter4k30_ofr_reliability
reds_inter4k30_ofr_sliding
reds_inter4k30_ofr_uniform
reds_inter4k30_reliability
reds_inter4k30_reliability_noimm
reds_inter4k30_reliability_rowwise
reds_inter4k30_sliding
reds_inter4k30_uniform
reds_inter4k_full_reliability
reds_inter4k_full_sliding
reds_reds4_sliding
reds_val30_gate_frame
reds_val30_gate_lam20_fifo
reds_val30_gate_lam5
reds_val30_reliability_v

目录结构：

<实验名>/
  <序列号>.mkv        # 该序列全部输出帧，FFV1 无损
  <序列号>/.done      # 原推理流程的完成标记
  manifest.json       # 帧数、原始 PNG 文件名、尺寸、逐序列体积

还原成 PNG 帧

归档为 bit-exact 无损——解码回的 PNG 与原始推理输出逐像素一致，可直接用于重算 PSNR/SSIM 等指标。manifest.json 记录了原始文件名，还原后命名与推理时完全一致。

# 需要 ffmpeg
python3 png_archive.py unpack <实验名> <输出目录>

# 或直接用 ffmpeg 解单个序列
ffmpeg -i 0025.mkv -pix_fmt rgb24 out/%08d.png

编码参数（ffv1 -level 3 -coder 1 -context 1 -g 1 -slices 4 -slicecrc 1 -pix_fmt rgb24）记录在每个 manifest 里。

注意：解码出的 PNG 文件字节大小与原始 PNG 不同（PNG 压缩级别不同），但像素数据完全一致。

实验设置

模型：FlashVSR v1（Tiny / Full）
测试集：REDS val30、REDS4、Inter4K、BSCV
KV 驱逐策略：sliding / uniform / headwise（FlashVSR 论文附录 B.3）、 reliability（自研无监督可靠性判据）、random、h2o，以及各类 gate 变体
输出分辨率：1280×640 RGB

指标结果（PSNR/SSIM/DOVER/时序一致性）另见仓库内的 metrics/。

引用

Downloads last month: 559

Total file size:

6.83 MB